prediction and see the results

open-mmlab / mmpose

OpenMMLab Pose Estimation Toolbox and Benchmark.

https://mmpose.readthedocs.io/en/latest/

Apache License 2.0

5.81k stars 1.24k forks source link

prediction and see the results #1841

Closed SylvainArd closed 1 year ago

SylvainArd commented 1 year ago

Hi I see in the documentation that we can do prediction on the complete dataset and generate a JSON of predictions or see the ground truth keypoints of the dataset but I see nowhere how to do the prediction on a new image and to see the predictions on the image predicted. The JSON file format generated by prediction is incompatible with those of the visualization. Furthermore I launched my training and after 140 epochs I have a loss of 2.0 and my keypoints are far of where it would be, the max score I have is 0.35. How to improve it please ?

ly015 commented 1 year ago

You can refer to the demo on how to visualize predictions on the image: https://github.com/open-mmlab/mmpose/tree/master/demo/docs. I am afraid that the JSON file can not be directly used for visualization.

About the model performance issue, could you please share the full config and the training log?

SylvainArd commented 1 year ago

I am not at home now so I will send it to you this evening in french hour

Envoyé de mon iPhone

Le 28 nov. 2022 à 04:04, Yining Li @.***> a écrit :

You can refer to the demo on how to visualize predictions on the image: https://github.com/open-mmlab/mmpose/tree/master/demo/docs. I am afraid that the JSON file can not be directly used for visualization.

About the model performance issue, could you please share the full config and the training log?

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.

SylvainArd commented 1 year ago

Hello, here is the entire folder of mmpose with all dataset and config files. I launched : python tools/train.py "C:\mmpose-master\configs\body\2d_kpt_sview_rgb_img\cid\coco\hrnet_w48_coco_512x512.py" --work-dir "output" thank you very much for your help ! Best regards

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le lun. 28 nov. 2022 à 10:27, Sylvain Ard @.***> a écrit :

I am not at home now so I will send it to you this evening in french hour

Envoyé de mon iPhone

Le 28 nov. 2022 à 04:04, Yining Li @.***> a écrit :

You can refer to the demo on how to visualize predictions on the image: https://github.com/open-mmlab/mmpose/tree/master/demo/docs. I am afraid that the JSON file can not be directly used for visualization.

About the model performance issue, could you please share the full config and the training log?

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1328468548, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR77G564KJA33LC2I63WKQOLRANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

I forgot the link : https://www.dropbox.com/sh/ttvjlp8iskznx4r/AAAvspg1II9D9VG41xgko5Rqa?dl=0 Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le lun. 28 nov. 2022 à 17:22, Sylvain Ard @.***> a écrit :

Hello, here is the entire folder of mmpose with all dataset and config files. I launched : python tools/train.py "C:\mmpose-master\configs\body\2d_kpt_sview_rgb_img\cid\coco\hrnet_w48_coco_512x512.py" --work-dir "output" thank you very much for your help ! Best regards

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le lun. 28 nov. 2022 à 10:27, Sylvain Ard @.***> a écrit :

I am not at home now so I will send it to you this evening in french hour

Envoyé de mon iPhone

Le 28 nov. 2022 à 04:04, Yining Li @.***> a écrit :

You can refer to the demo on how to visualize predictions on the image: https://github.com/open-mmlab/mmpose/tree/master/demo/docs. I am afraid that the JSON file can not be directly used for visualization.

About the model performance issue, could you please share the full config and the training log?

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1328468548, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR77G564KJA33LC2I63WKQOLRANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

up please

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le lun. 28 nov. 2022 à 17:36, Sylvain Ard @.***> a écrit :

I forgot the link : https://www.dropbox.com/sh/ttvjlp8iskznx4r/AAAvspg1II9D9VG41xgko5Rqa?dl=0 Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le lun. 28 nov. 2022 à 17:22, Sylvain Ard @.***> a écrit :

Hello, here is the entire folder of mmpose with all dataset and config files. I launched : python tools/train.py "C:\mmpose-master\configs\body\2d_kpt_sview_rgb_img\cid\coco\hrnet_w48_coco_512x512.py" --work-dir "output" thank you very much for your help ! Best regards

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le lun. 28 nov. 2022 à 10:27, Sylvain Ard @.***> a écrit :

I am not at home now so I will send it to you this evening in french hour

Envoyé de mon iPhone

Le 28 nov. 2022 à 04:04, Yining Li @.***> a écrit :

You can refer to the demo on how to visualize predictions on the image: https://github.com/open-mmlab/mmpose/tree/master/demo/docs. I am afraid that the JSON file can not be directly used for visualization.

About the model performance issue, could you please share the full config and the training log?

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1328468548, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR77G564KJA33LC2I63WKQOLRANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

Have I just to do more iterations .? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le mer. 30 nov. 2022 à 09:49, Sylvain Ard @.***> a écrit :

up please

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le lun. 28 nov. 2022 à 17:36, Sylvain Ard @.***> a écrit :

I forgot the link : https://www.dropbox.com/sh/ttvjlp8iskznx4r/AAAvspg1II9D9VG41xgko5Rqa?dl=0 Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le lun. 28 nov. 2022 à 17:22, Sylvain Ard @.***> a écrit :

Hello, here is the entire folder of mmpose with all dataset and config files. I launched : python tools/train.py "C:\mmpose-master\configs\body\2d_kpt_sview_rgb_img\cid\coco\hrnet_w48_coco_512x512.py" --work-dir "output" thank you very much for your help ! Best regards

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le lun. 28 nov. 2022 à 10:27, Sylvain Ard @.***> a écrit :

I am not at home now so I will send it to you this evening in french hour

Envoyé de mon iPhone

Le 28 nov. 2022 à 04:04, Yining Li @.***> a écrit :

You can refer to the demo on how to visualize predictions on the image: https://github.com/open-mmlab/mmpose/tree/master/demo/docs. I am afraid that the JSON file can not be directly used for visualization.

About the model performance issue, could you please share the full config and the training log?

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1328468548, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR77G564KJA33LC2I63WKQOLRANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

ly015 commented 1 year ago

Hi, I have taken a look at the config and I have a few questions:

The training and testing dataset uses the same annotation file. Is it intended to use the same data for both training and testing?
The training batch size is set as 1 (assume 1 GPU is used) which may be too small.
There are lots of log files in the output folder and I am not sure which one to check.

ly015 commented 1 year ago

BTW, I would suggest starting from a simpler model like a top-down heatmap model instead of CID. The former should be easier to tune on your own data. Maybe HRNet is what you will be interested in. Of course, you would need to train a detection model to extract object bbox for a top-down keypoint model.

SylvainArd commented 1 year ago

the logs are very similar there are two warnings but I don't understand these :

2022-11-25 14:48:28,267 - mmpose - WARNING - The model and loaded state dict do not match exactly

unexpected key in source state_dict: head.0.0.0.conv1.weight, head.0.0.0.bn1.weight, head.0.0.0.bn1.bias, head.0.0.0.bn1.running_mean, head.0.0.0.bn1.running_var, head.0.0.0.conv2.weight, head.0.0.0.bn2.weight, head.0.0.0.bn2.bias, head.0.0.0.bn2.running_mean, head.0.0.0.bn2.running_var, head.0.0.0.conv3.weight, head.0.0.0.bn3.weight, head.0.0.0.bn3.bias, head.0.0.0.bn3.running_mean, head.0.0.0.bn3.running_var, head.0.0.0.downsample.0.weight, head.0.0.0.downsample.1.weight, head.0.0.0.downsample.1.bias, head.0.0.0.downsample.1.running_mean, head.0.0.0.downsample.1.running_var, head.0.1.0.conv1.weight, head.0.1.0.bn1.weight, head.0.1.0.bn1.bias, head.0.1.0.bn1.running_mean, head.0.1.0.bn1.running_var, head.0.1.0.conv2.weight, head.0.1.0.bn2.weight, head.0.1.0.bn2.bias, head.0.1.0.bn2.running_mean, head.0.1.0.bn2.running_var, head.0.1.0.conv3.weight, head.0.1.0.bn3.weight, head.0.1.0.bn3.bias, head.0.1.0.bn3.running_mean, head.0.1.0.bn3.running_var, head.0.1.0.downsample.0.weight, head.0.1.0.downsample.1.weight, head.0.1.0.downsample.1.bias, head.0.1.0.downsample.1.running_mean, head.0.1.0.downsample.1.running_var, head.0.2.0.conv1.weight, head.0.2.0.bn1.weight, head.0.2.0.bn1.bias, head.0.2.0.bn1.running_mean, head.0.2.0.bn1.running_var, head.0.2.0.conv2.weight, head.0.2.0.bn2.weight, head.0.2.0.bn2.bias, head.0.2.0.bn2.running_mean, head.0.2.0.bn2.running_var, head.0.2.0.conv3.weight, head.0.2.0.bn3.weight, head.0.2.0.bn3.bias, head.0.2.0.bn3.running_mean, head.0.2.0.bn3.running_var, head.0.2.0.downsample.0.weight, head.0.2.0.downsample.1.weight, head.0.2.0.downsample.1.bias, head.0.2.0.downsample.1.running_mean, head.0.2.0.downsample.1.running_var, head.1.0.0.conv1.weight, head.1.0.0.bn1.weight, head.1.0.0.bn1.bias, head.1.0.0.bn1.running_mean, head.1.0.0.bn1.running_var, head.1.0.0.conv2.weight, head.1.0.0.bn2.weight, head.1.0.0.bn2.bias, head.1.0.0.bn2.running_mean, head.1.0.0.bn2.running_var, head.1.0.0.conv3.weight, head.1.0.0.bn3.weight, head.1.0.0.bn3.bias, head.1.0.0.bn3.running_mean, head.1.0.0.bn3.running_var, head.1.0.0.downsample.0.weight, head.1.0.0.downsample.1.weight, head.1.0.0.downsample.1.bias, head.1.0.0.downsample.1.running_mean, head.1.0.0.downsample.1.running_var, head.1.1.0.conv1.weight, head.1.1.0.bn1.weight, head.1.1.0.bn1.bias, head.1.1.0.bn1.running_mean, head.1.1.0.bn1.running_var, head.1.1.0.conv2.weight, head.1.1.0.bn2.weight, head.1.1.0.bn2.bias, head.1.1.0.bn2.running_mean, head.1.1.0.bn2.running_var, head.1.1.0.conv3.weight, head.1.1.0.bn3.weight, head.1.1.0.bn3.bias, head.1.1.0.bn3.running_mean, head.1.1.0.bn3.running_var, head.1.1.0.downsample.0.weight, head.1.1.0.downsample.1.weight, head.1.1.0.downsample.1.bias, head.1.1.0.downsample.1.running_mean, head.1.1.0.downsample.1.running_var, head.2.0.0.conv1.weight, head.2.0.0.bn1.weight, head.2.0.0.bn1.bias, head.2.0.0.bn1.running_mean, head.2.0.0.bn1.running_var, head.2.0.0.conv2.weight, head.2.0.0.bn2.weight, head.2.0.0.bn2.bias, head.2.0.0.bn2.running_mean, head.2.0.0.bn2.running_var, head.2.0.0.conv3.weight, head.2.0.0.bn3.weight, head.2.0.0.bn3.bias, head.2.0.0.bn3.running_mean, head.2.0.0.bn3.running_var, head.2.0.0.downsample.0.weight, head.2.0.0.downsample.1.weight, head.2.0.0.downsample.1.bias, head.2.0.0.downsample.1.running_mean, head.2.0.0.downsample.1.running_var, head.3.0.0.conv1.weight, head.3.0.0.bn1.weight, head.3.0.0.bn1.bias, head.3.0.0.bn1.running_mean, head.3.0.0.bn1.running_var, head.3.0.0.conv2.weight, head.3.0.0.bn2.weight, head.3.0.0.bn2.bias, head.3.0.0.bn2.running_mean, head.3.0.0.bn2.running_var, head.3.0.0.conv3.weight, head.3.0.0.bn3.weight, head.3.0.0.bn3.bias, head.3.0.0.bn3.running_mean, head.3.0.0.bn3.running_var, head.3.0.0.downsample.0.weight, head.3.0.0.downsample.1.weight, head.3.0.0.downsample.1.bias, head.3.0.0.downsample.1.running_mean, head.3.0.0.downsample.1.running_var, fc.weight, fc.bias

The maximum bach size I can do with 1024 image size is 3 Sylvain Ard 054950772 4 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:32, Yining Li @.***> a écrit :

BTW, I would suggest starting from a simpler model like a top-down heatmap model instead of CID. The former should be easier to tune on your own data. Maybe HRNet https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md is what you may be interested in. Of course, you would need to train a detection model to extract object bbox for a top-down keypoint model.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333478955, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR4B3EALV7A6GPNXNILWLBWDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

I don't think the model you suggest me is simpler than cid, the parameters are the same Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:45, Sylvain Ard @.***> a écrit :

the logs are very similar there are two warnings but I don't understand these :

2022-11-25 14:48:28,267 - mmpose - WARNING - The model and loaded state dict do not match exactly

unexpected key in source state_dict: head.0.0.0.conv1.weight, head.0.0.0.bn1.weight, head.0.0.0.bn1.bias, head.0.0.0.bn1.running_mean, head.0.0.0.bn1.running_var, head.0.0.0.conv2.weight, head.0.0.0.bn2.weight, head.0.0.0.bn2.bias, head.0.0.0.bn2.running_mean, head.0.0.0.bn2.running_var, head.0.0.0.conv3.weight, head.0.0.0.bn3.weight, head.0.0.0.bn3.bias, head.0.0.0.bn3.running_mean, head.0.0.0.bn3.running_var, head.0.0.0.downsample.0.weight, head.0.0.0.downsample.1.weight, head.0.0.0.downsample.1.bias, head.0.0.0.downsample.1.running_mean, head.0.0.0.downsample.1.running_var, head.0.1.0.conv1.weight, head.0.1.0.bn1.weight, head.0.1.0.bn1.bias, head.0.1.0.bn1.running_mean, head.0.1.0.bn1.running_var, head.0.1.0.conv2.weight, head.0.1.0.bn2.weight, head.0.1.0.bn2.bias, head.0.1.0.bn2.running_mean, head.0.1.0.bn2.running_var, head.0.1.0.conv3.weight, head.0.1.0.bn3.weight, head.0.1.0.bn3.bias, head.0.1.0.bn3.running_mean, head.0.1.0.bn3.running_var, head.0.1.0.downsample.0.weight, head.0.1.0.downsample.1.weight, head.0.1.0.downsample.1.bias, head.0.1.0.downsample.1.running_mean, head.0.1.0.downsample.1.running_var, head.0.2.0.conv1.weight, head.0.2.0.bn1.weight, head.0.2.0.bn1.bias, head.0.2.0.bn1.running_mean, head.0.2.0.bn1.running_var, head.0.2.0.conv2.weight, head.0.2.0.bn2.weight, head.0.2.0.bn2.bias, head.0.2.0.bn2.running_mean, head.0.2.0.bn2.running_var, head.0.2.0.conv3.weight, head.0.2.0.bn3.weight, head.0.2.0.bn3.bias, head.0.2.0.bn3.running_mean, head.0.2.0.bn3.running_var, head.0.2.0.downsample.0.weight, head.0.2.0.downsample.1.weight, head.0.2.0.downsample.1.bias, head.0.2.0.downsample.1.running_mean, head.0.2.0.downsample.1.running_var, head.1.0.0.conv1.weight, head.1.0.0.bn1.weight, head.1.0.0.bn1.bias, head.1.0.0.bn1.running_mean, head.1.0.0.bn1.running_var, head.1.0.0.conv2.weight, head.1.0.0.bn2.weight, head.1.0.0.bn2.bias, head.1.0.0.bn2.running_mean, head.1.0.0.bn2.running_var, head.1.0.0.conv3.weight, head.1.0.0.bn3.weight, head.1.0.0.bn3.bias, head.1.0.0.bn3.running_mean, head.1.0.0.bn3.running_var, head.1.0.0.downsample.0.weight, head.1.0.0.downsample.1.weight, head.1.0.0.downsample.1.bias, head.1.0.0.downsample.1.running_mean, head.1.0.0.downsample.1.running_var, head.1.1.0.conv1.weight, head.1.1.0.bn1.weight, head.1.1.0.bn1.bias, head.1.1.0.bn1.running_mean, head.1.1.0.bn1.running_var, head.1.1.0.conv2.weight, head.1.1.0.bn2.weight, head.1.1.0.bn2.bias, head.1.1.0.bn2.running_mean, head.1.1.0.bn2.running_var, head.1.1.0.conv3.weight, head.1.1.0.bn3.weight, head.1.1.0.bn3.bias, head.1.1.0.bn3.running_mean, head.1.1.0.bn3.running_var, head.1.1.0.downsample.0.weight, head.1.1.0.downsample.1.weight, head.1.1.0.downsample.1.bias, head.1.1.0.downsample.1.running_mean, head.1.1.0.downsample.1.running_var, head.2.0.0.conv1.weight, head.2.0.0.bn1.weight, head.2.0.0.bn1.bias, head.2.0.0.bn1.running_mean, head.2.0.0.bn1.running_var, head.2.0.0.conv2.weight, head.2.0.0.bn2.weight, head.2.0.0.bn2.bias, head.2.0.0.bn2.running_mean, head.2.0.0.bn2.running_var, head.2.0.0.conv3.weight, head.2.0.0.bn3.weight, head.2.0.0.bn3.bias, head.2.0.0.bn3.running_mean, head.2.0.0.bn3.running_var, head.2.0.0.downsample.0.weight, head.2.0.0.downsample.1.weight, head.2.0.0.downsample.1.bias, head.2.0.0.downsample.1.running_mean, head.2.0.0.downsample.1.running_var, head.3.0.0.conv1.weight, head.3.0.0.bn1.weight, head.3.0.0.bn1.bias, head.3.0.0.bn1.running_mean, head.3.0.0.bn1.running_var, head.3.0.0.conv2.weight, head.3.0.0.bn2.weight, head.3.0.0.bn2.bias, head.3.0.0.bn2.running_mean, head.3.0.0.bn2.running_var, head.3.0.0.conv3.weight, head.3.0.0.bn3.weight, head.3.0.0.bn3.bias, head.3.0.0.bn3.running_mean, head.3.0.0.bn3.running_var, head.3.0.0.downsample.0.weight, head.3.0.0.downsample.1.weight, head.3.0.0.downsample.1.bias, head.3.0.0.downsample.1.running_mean, head.3.0.0.downsample.1.running_var, fc.weight, fc.bias

The maximum bach size I can do with 1024 image size is 3 Sylvain Ard 054950772 4 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:32, Yining Li @.***> a écrit :

BTW, I would suggest starting from a simpler model like a top-down heatmap model instead of CID. The former should be easier to tune on your own data. Maybe HRNet https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md is what you may be interested in. Of course, you would need to train a detection model to extract object bbox for a top-down keypoint model.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333478955, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR4B3EALV7A6GPNXNILWLBWDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

when I do batch_size=3 the training does not advance Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:48, Sylvain Ard @.***> a écrit :

I don't think the model you suggest me is simpler than cid, the parameters are the same Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:45, Sylvain Ard @.***> a écrit :

the logs are very similar there are two warnings but I don't understand these :

2022-11-25 14:48:28,267 - mmpose - WARNING - The model and loaded state dict do not match exactly

unexpected key in source state_dict: head.0.0.0.conv1.weight, head.0.0.0.bn1.weight, head.0.0.0.bn1.bias, head.0.0.0.bn1.running_mean, head.0.0.0.bn1.running_var, head.0.0.0.conv2.weight, head.0.0.0.bn2.weight, head.0.0.0.bn2.bias, head.0.0.0.bn2.running_mean, head.0.0.0.bn2.running_var, head.0.0.0.conv3.weight, head.0.0.0.bn3.weight, head.0.0.0.bn3.bias, head.0.0.0.bn3.running_mean, head.0.0.0.bn3.running_var, head.0.0.0.downsample.0.weight, head.0.0.0.downsample.1.weight, head.0.0.0.downsample.1.bias, head.0.0.0.downsample.1.running_mean, head.0.0.0.downsample.1.running_var, head.0.1.0.conv1.weight, head.0.1.0.bn1.weight, head.0.1.0.bn1.bias, head.0.1.0.bn1.running_mean, head.0.1.0.bn1.running_var, head.0.1.0.conv2.weight, head.0.1.0.bn2.weight, head.0.1.0.bn2.bias, head.0.1.0.bn2.running_mean, head.0.1.0.bn2.running_var, head.0.1.0.conv3.weight, head.0.1.0.bn3.weight, head.0.1.0.bn3.bias, head.0.1.0.bn3.running_mean, head.0.1.0.bn3.running_var, head.0.1.0.downsample.0.weight, head.0.1.0.downsample.1.weight, head.0.1.0.downsample.1.bias, head.0.1.0.downsample.1.running_mean, head.0.1.0.downsample.1.running_var, head.0.2.0.conv1.weight, head.0.2.0.bn1.weight, head.0.2.0.bn1.bias, head.0.2.0.bn1.running_mean, head.0.2.0.bn1.running_var, head.0.2.0.conv2.weight, head.0.2.0.bn2.weight, head.0.2.0.bn2.bias, head.0.2.0.bn2.running_mean, head.0.2.0.bn2.running_var, head.0.2.0.conv3.weight, head.0.2.0.bn3.weight, head.0.2.0.bn3.bias, head.0.2.0.bn3.running_mean, head.0.2.0.bn3.running_var, head.0.2.0.downsample.0.weight, head.0.2.0.downsample.1.weight, head.0.2.0.downsample.1.bias, head.0.2.0.downsample.1.running_mean, head.0.2.0.downsample.1.running_var, head.1.0.0.conv1.weight, head.1.0.0.bn1.weight, head.1.0.0.bn1.bias, head.1.0.0.bn1.running_mean, head.1.0.0.bn1.running_var, head.1.0.0.conv2.weight, head.1.0.0.bn2.weight, head.1.0.0.bn2.bias, head.1.0.0.bn2.running_mean, head.1.0.0.bn2.running_var, head.1.0.0.conv3.weight, head.1.0.0.bn3.weight, head.1.0.0.bn3.bias, head.1.0.0.bn3.running_mean, head.1.0.0.bn3.running_var, head.1.0.0.downsample.0.weight, head.1.0.0.downsample.1.weight, head.1.0.0.downsample.1.bias, head.1.0.0.downsample.1.running_mean, head.1.0.0.downsample.1.running_var, head.1.1.0.conv1.weight, head.1.1.0.bn1.weight, head.1.1.0.bn1.bias, head.1.1.0.bn1.running_mean, head.1.1.0.bn1.running_var, head.1.1.0.conv2.weight, head.1.1.0.bn2.weight, head.1.1.0.bn2.bias, head.1.1.0.bn2.running_mean, head.1.1.0.bn2.running_var, head.1.1.0.conv3.weight, head.1.1.0.bn3.weight, head.1.1.0.bn3.bias, head.1.1.0.bn3.running_mean, head.1.1.0.bn3.running_var, head.1.1.0.downsample.0.weight, head.1.1.0.downsample.1.weight, head.1.1.0.downsample.1.bias, head.1.1.0.downsample.1.running_mean, head.1.1.0.downsample.1.running_var, head.2.0.0.conv1.weight, head.2.0.0.bn1.weight, head.2.0.0.bn1.bias, head.2.0.0.bn1.running_mean, head.2.0.0.bn1.running_var, head.2.0.0.conv2.weight, head.2.0.0.bn2.weight, head.2.0.0.bn2.bias, head.2.0.0.bn2.running_mean, head.2.0.0.bn2.running_var, head.2.0.0.conv3.weight, head.2.0.0.bn3.weight, head.2.0.0.bn3.bias, head.2.0.0.bn3.running_mean, head.2.0.0.bn3.running_var, head.2.0.0.downsample.0.weight, head.2.0.0.downsample.1.weight, head.2.0.0.downsample.1.bias, head.2.0.0.downsample.1.running_mean, head.2.0.0.downsample.1.running_var, head.3.0.0.conv1.weight, head.3.0.0.bn1.weight, head.3.0.0.bn1.bias, head.3.0.0.bn1.running_mean, head.3.0.0.bn1.running_var, head.3.0.0.conv2.weight, head.3.0.0.bn2.weight, head.3.0.0.bn2.bias, head.3.0.0.bn2.running_mean, head.3.0.0.bn2.running_var, head.3.0.0.conv3.weight, head.3.0.0.bn3.weight, head.3.0.0.bn3.bias, head.3.0.0.bn3.running_mean, head.3.0.0.bn3.running_var, head.3.0.0.downsample.0.weight, head.3.0.0.downsample.1.weight, head.3.0.0.downsample.1.bias, head.3.0.0.downsample.1.running_mean, head.3.0.0.downsample.1.running_var, fc.weight, fc.bias

The maximum bach size I can do with 1024 image size is 3 Sylvain Ard 054950772 4 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:32, Yining Li @.***> a écrit :

BTW, I would suggest starting from a simpler model like a top-down heatmap model instead of CID. The former should be easier to tune on your own data. Maybe HRNet https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md is what you may be interested in. Of course, you would need to train a detection model to extract object bbox for a top-down keypoint model.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333478955, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR4B3EALV7A6GPNXNILWLBWDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

The training and testing dataset uses the same annotation file. Is it intended to use the same data for both training and testing? >> yes
The training batch size is set as 1 (assume 1 GPU is used) which may be too small. >> I tried 3 but I don't see the progression, and I don't know if it advance or not
There are lots of log files in the output folder and I am not sure which one to check.

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:57, Sylvain Ard @.***> a écrit :

when I do batch_size=3 the training does not advance Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:48, Sylvain Ard @.***> a écrit :

I don't think the model you suggest me is simpler than cid, the parameters are the same Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:45, Sylvain Ard @.***> a écrit :

the logs are very similar there are two warnings but I don't understand these :

2022-11-25 14:48:28,267 - mmpose - WARNING - The model and loaded state dict do not match exactly

unexpected key in source state_dict: head.0.0.0.conv1.weight, head.0.0.0.bn1.weight, head.0.0.0.bn1.bias, head.0.0.0.bn1.running_mean, head.0.0.0.bn1.running_var, head.0.0.0.conv2.weight, head.0.0.0.bn2.weight, head.0.0.0.bn2.bias, head.0.0.0.bn2.running_mean, head.0.0.0.bn2.running_var, head.0.0.0.conv3.weight, head.0.0.0.bn3.weight, head.0.0.0.bn3.bias, head.0.0.0.bn3.running_mean, head.0.0.0.bn3.running_var, head.0.0.0.downsample.0.weight, head.0.0.0.downsample.1.weight, head.0.0.0.downsample.1.bias, head.0.0.0.downsample.1.running_mean, head.0.0.0.downsample.1.running_var, head.0.1.0.conv1.weight, head.0.1.0.bn1.weight, head.0.1.0.bn1.bias, head.0.1.0.bn1.running_mean, head.0.1.0.bn1.running_var, head.0.1.0.conv2.weight, head.0.1.0.bn2.weight, head.0.1.0.bn2.bias, head.0.1.0.bn2.running_mean, head.0.1.0.bn2.running_var, head.0.1.0.conv3.weight, head.0.1.0.bn3.weight, head.0.1.0.bn3.bias, head.0.1.0.bn3.running_mean, head.0.1.0.bn3.running_var, head.0.1.0.downsample.0.weight, head.0.1.0.downsample.1.weight, head.0.1.0.downsample.1.bias, head.0.1.0.downsample.1.running_mean, head.0.1.0.downsample.1.running_var, head.0.2.0.conv1.weight, head.0.2.0.bn1.weight, head.0.2.0.bn1.bias, head.0.2.0.bn1.running_mean, head.0.2.0.bn1.running_var, head.0.2.0.conv2.weight, head.0.2.0.bn2.weight, head.0.2.0.bn2.bias, head.0.2.0.bn2.running_mean, head.0.2.0.bn2.running_var, head.0.2.0.conv3.weight, head.0.2.0.bn3.weight, head.0.2.0.bn3.bias, head.0.2.0.bn3.running_mean, head.0.2.0.bn3.running_var, head.0.2.0.downsample.0.weight, head.0.2.0.downsample.1.weight, head.0.2.0.downsample.1.bias, head.0.2.0.downsample.1.running_mean, head.0.2.0.downsample.1.running_var, head.1.0.0.conv1.weight, head.1.0.0.bn1.weight, head.1.0.0.bn1.bias, head.1.0.0.bn1.running_mean, head.1.0.0.bn1.running_var, head.1.0.0.conv2.weight, head.1.0.0.bn2.weight, head.1.0.0.bn2.bias, head.1.0.0.bn2.running_mean, head.1.0.0.bn2.running_var, head.1.0.0.conv3.weight, head.1.0.0.bn3.weight, head.1.0.0.bn3.bias, head.1.0.0.bn3.running_mean, head.1.0.0.bn3.running_var, head.1.0.0.downsample.0.weight, head.1.0.0.downsample.1.weight, head.1.0.0.downsample.1.bias, head.1.0.0.downsample.1.running_mean, head.1.0.0.downsample.1.running_var, head.1.1.0.conv1.weight, head.1.1.0.bn1.weight, head.1.1.0.bn1.bias, head.1.1.0.bn1.running_mean, head.1.1.0.bn1.running_var, head.1.1.0.conv2.weight, head.1.1.0.bn2.weight, head.1.1.0.bn2.bias, head.1.1.0.bn2.running_mean, head.1.1.0.bn2.running_var, head.1.1.0.conv3.weight, head.1.1.0.bn3.weight, head.1.1.0.bn3.bias, head.1.1.0.bn3.running_mean, head.1.1.0.bn3.running_var, head.1.1.0.downsample.0.weight, head.1.1.0.downsample.1.weight, head.1.1.0.downsample.1.bias, head.1.1.0.downsample.1.running_mean, head.1.1.0.downsample.1.running_var, head.2.0.0.conv1.weight, head.2.0.0.bn1.weight, head.2.0.0.bn1.bias, head.2.0.0.bn1.running_mean, head.2.0.0.bn1.running_var, head.2.0.0.conv2.weight, head.2.0.0.bn2.weight, head.2.0.0.bn2.bias, head.2.0.0.bn2.running_mean, head.2.0.0.bn2.running_var, head.2.0.0.conv3.weight, head.2.0.0.bn3.weight, head.2.0.0.bn3.bias, head.2.0.0.bn3.running_mean, head.2.0.0.bn3.running_var, head.2.0.0.downsample.0.weight, head.2.0.0.downsample.1.weight, head.2.0.0.downsample.1.bias, head.2.0.0.downsample.1.running_mean, head.2.0.0.downsample.1.running_var, head.3.0.0.conv1.weight, head.3.0.0.bn1.weight, head.3.0.0.bn1.bias, head.3.0.0.bn1.running_mean, head.3.0.0.bn1.running_var, head.3.0.0.conv2.weight, head.3.0.0.bn2.weight, head.3.0.0.bn2.bias, head.3.0.0.bn2.running_mean, head.3.0.0.bn2.running_var, head.3.0.0.conv3.weight, head.3.0.0.bn3.weight, head.3.0.0.bn3.bias, head.3.0.0.bn3.running_mean, head.3.0.0.bn3.running_var, head.3.0.0.downsample.0.weight, head.3.0.0.downsample.1.weight, head.3.0.0.downsample.1.bias, head.3.0.0.downsample.1.running_mean, head.3.0.0.downsample.1.running_var, fc.weight, fc.bias

The maximum bach size I can do with 1024 image size is 3 Sylvain Ard 054950772 4 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:32, Yining Li @.***> a écrit :

BTW, I would suggest starting from a simpler model like a top-down heatmap model instead of CID. The former should be easier to tune on your own data. Maybe HRNet https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md is what you may be interested in. Of course, you would need to train a detection model to extract object bbox for a top-down keypoint model.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333478955, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR4B3EALV7A6GPNXNILWLBWDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

I think the only part where we can do something is :

backbone=dict( type='HRNet', in_channels=3, extra=dict( stage1=dict( num_modules=1, num_branches=1, block='BOTTLENECK', num_blocks=(4, ), num_channels=(64, )), stage2=dict( num_modules=1, num_branches=2, block='BASIC', num_blocks=(4, 4), num_channels=(48, 96)), stage3=dict( num_modules=4, num_branches=3, block='BASIC', num_blocks=(4, 4, 4), num_channels=(48, 96, 192)), stage4=dict( num_modules=3, num_branches=4, block='BASIC', num_blocks=(4, 4, 4, 4), num_channels=(48, 96, 192, 384), multiscale_output=True)), ),

but I don't know what, an idea please ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 11:04, Sylvain Ard @.***> a écrit :

The training and testing dataset uses the same annotation file. Is it intended to use the same data for both training and testing? >> yes

The training batch size is set as 1 (assume 1 GPU is used) which may be too small. >> I tried 3 but I don't see the progression, and I don't know if it advance or not

There are lots of log files in the output folder and I am not sure which one to check.

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:57, Sylvain Ard @.***> a écrit :

when I do batch_size=3 the training does not advance Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:48, Sylvain Ard @.***> a écrit :

I don't think the model you suggest me is simpler than cid, the parameters are the same Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:45, Sylvain Ard @.***> a écrit :

the logs are very similar there are two warnings but I don't understand these :

2022-11-25 14:48:28,267 - mmpose - WARNING - The model and loaded state dict do not match exactly

unexpected key in source state_dict: head.0.0.0.conv1.weight, head.0.0.0.bn1.weight, head.0.0.0.bn1.bias, head.0.0.0.bn1.running_mean, head.0.0.0.bn1.running_var, head.0.0.0.conv2.weight, head.0.0.0.bn2.weight, head.0.0.0.bn2.bias, head.0.0.0.bn2.running_mean, head.0.0.0.bn2.running_var, head.0.0.0.conv3.weight, head.0.0.0.bn3.weight, head.0.0.0.bn3.bias, head.0.0.0.bn3.running_mean, head.0.0.0.bn3.running_var, head.0.0.0.downsample.0.weight, head.0.0.0.downsample.1.weight, head.0.0.0.downsample.1.bias, head.0.0.0.downsample.1.running_mean, head.0.0.0.downsample.1.running_var, head.0.1.0.conv1.weight, head.0.1.0.bn1.weight, head.0.1.0.bn1.bias, head.0.1.0.bn1.running_mean, head.0.1.0.bn1.running_var, head.0.1.0.conv2.weight, head.0.1.0.bn2.weight, head.0.1.0.bn2.bias, head.0.1.0.bn2.running_mean, head.0.1.0.bn2.running_var, head.0.1.0.conv3.weight, head.0.1.0.bn3.weight, head.0.1.0.bn3.bias, head.0.1.0.bn3.running_mean, head.0.1.0.bn3.running_var, head.0.1.0.downsample.0.weight, head.0.1.0.downsample.1.weight, head.0.1.0.downsample.1.bias, head.0.1.0.downsample.1.running_mean, head.0.1.0.downsample.1.running_var, head.0.2.0.conv1.weight, head.0.2.0.bn1.weight, head.0.2.0.bn1.bias, head.0.2.0.bn1.running_mean, head.0.2.0.bn1.running_var, head.0.2.0.conv2.weight, head.0.2.0.bn2.weight, head.0.2.0.bn2.bias, head.0.2.0.bn2.running_mean, head.0.2.0.bn2.running_var, head.0.2.0.conv3.weight, head.0.2.0.bn3.weight, head.0.2.0.bn3.bias, head.0.2.0.bn3.running_mean, head.0.2.0.bn3.running_var, head.0.2.0.downsample.0.weight, head.0.2.0.downsample.1.weight, head.0.2.0.downsample.1.bias, head.0.2.0.downsample.1.running_mean, head.0.2.0.downsample.1.running_var, head.1.0.0.conv1.weight, head.1.0.0.bn1.weight, head.1.0.0.bn1.bias, head.1.0.0.bn1.running_mean, head.1.0.0.bn1.running_var, head.1.0.0.conv2.weight, head.1.0.0.bn2.weight, head.1.0.0.bn2.bias, head.1.0.0.bn2.running_mean, head.1.0.0.bn2.running_var, head.1.0.0.conv3.weight, head.1.0.0.bn3.weight, head.1.0.0.bn3.bias, head.1.0.0.bn3.running_mean, head.1.0.0.bn3.running_var, head.1.0.0.downsample.0.weight, head.1.0.0.downsample.1.weight, head.1.0.0.downsample.1.bias, head.1.0.0.downsample.1.running_mean, head.1.0.0.downsample.1.running_var, head.1.1.0.conv1.weight, head.1.1.0.bn1.weight, head.1.1.0.bn1.bias, head.1.1.0.bn1.running_mean, head.1.1.0.bn1.running_var, head.1.1.0.conv2.weight, head.1.1.0.bn2.weight, head.1.1.0.bn2.bias, head.1.1.0.bn2.running_mean, head.1.1.0.bn2.running_var, head.1.1.0.conv3.weight, head.1.1.0.bn3.weight, head.1.1.0.bn3.bias, head.1.1.0.bn3.running_mean, head.1.1.0.bn3.running_var, head.1.1.0.downsample.0.weight, head.1.1.0.downsample.1.weight, head.1.1.0.downsample.1.bias, head.1.1.0.downsample.1.running_mean, head.1.1.0.downsample.1.running_var, head.2.0.0.conv1.weight, head.2.0.0.bn1.weight, head.2.0.0.bn1.bias, head.2.0.0.bn1.running_mean, head.2.0.0.bn1.running_var, head.2.0.0.conv2.weight, head.2.0.0.bn2.weight, head.2.0.0.bn2.bias, head.2.0.0.bn2.running_mean, head.2.0.0.bn2.running_var, head.2.0.0.conv3.weight, head.2.0.0.bn3.weight, head.2.0.0.bn3.bias, head.2.0.0.bn3.running_mean, head.2.0.0.bn3.running_var, head.2.0.0.downsample.0.weight, head.2.0.0.downsample.1.weight, head.2.0.0.downsample.1.bias, head.2.0.0.downsample.1.running_mean, head.2.0.0.downsample.1.running_var, head.3.0.0.conv1.weight, head.3.0.0.bn1.weight, head.3.0.0.bn1.bias, head.3.0.0.bn1.running_mean, head.3.0.0.bn1.running_var, head.3.0.0.conv2.weight, head.3.0.0.bn2.weight, head.3.0.0.bn2.bias, head.3.0.0.bn2.running_mean, head.3.0.0.bn2.running_var, head.3.0.0.conv3.weight, head.3.0.0.bn3.weight, head.3.0.0.bn3.bias, head.3.0.0.bn3.running_mean, head.3.0.0.bn3.running_var, head.3.0.0.downsample.0.weight, head.3.0.0.downsample.1.weight, head.3.0.0.downsample.1.bias, head.3.0.0.downsample.1.running_mean, head.3.0.0.downsample.1.running_var, fc.weight, fc.bias

The maximum bach size I can do with 1024 image size is 3 Sylvain Ard 054950772 4 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:32, Yining Li @.***> a écrit :

BTW, I would suggest starting from a simpler model like a top-down heatmap model instead of CID. The former should be easier to tune on your own data. Maybe HRNet https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md is what you may be interested in. Of course, you would need to train a detection model to extract object bbox for a top-down keypoint model.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333478955, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR4B3EALV7A6GPNXNILWLBWDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

and if we change the learning rate ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 11:23, Sylvain Ard @.***> a écrit :

I think the only part where we can do something is :

backbone=dict( type='HRNet', in_channels=3, extra=dict( stage1=dict( num_modules=1, num_branches=1, block='BOTTLENECK', num_blocks=(4, ), num_channels=(64, )), stage2=dict( num_modules=1, num_branches=2, block='BASIC', num_blocks=(4, 4), num_channels=(48, 96)), stage3=dict( num_modules=4, num_branches=3, block='BASIC', num_blocks=(4, 4, 4), num_channels=(48, 96, 192)), stage4=dict( num_modules=3, num_branches=4, block='BASIC', num_blocks=(4, 4, 4, 4), num_channels=(48, 96, 192, 384), multiscale_output=True)), ),

but I don't know what, an idea please ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 11:04, Sylvain Ard @.***> a écrit :

The training and testing dataset uses the same annotation file. Is it intended to use the same data for both training and testing? >> yes

The training batch size is set as 1 (assume 1 GPU is used) which may be too small. >> I tried 3 but I don't see the progression, and I don't know if it advance or not

There are lots of log files in the output folder and I am not sure which one to check.

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:57, Sylvain Ard @.***> a écrit :

when I do batch_size=3 the training does not advance Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:48, Sylvain Ard @.***> a écrit :

I don't think the model you suggest me is simpler than cid, the parameters are the same Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:45, Sylvain Ard @.***> a écrit :

the logs are very similar there are two warnings but I don't understand these :

2022-11-25 14:48:28,267 - mmpose - WARNING - The model and loaded state dict do not match exactly

unexpected key in source state_dict: head.0.0.0.conv1.weight, head.0.0.0.bn1.weight, head.0.0.0.bn1.bias, head.0.0.0.bn1.running_mean, head.0.0.0.bn1.running_var, head.0.0.0.conv2.weight, head.0.0.0.bn2.weight, head.0.0.0.bn2.bias, head.0.0.0.bn2.running_mean, head.0.0.0.bn2.running_var, head.0.0.0.conv3.weight, head.0.0.0.bn3.weight, head.0.0.0.bn3.bias, head.0.0.0.bn3.running_mean, head.0.0.0.bn3.running_var, head.0.0.0.downsample.0.weight, head.0.0.0.downsample.1.weight, head.0.0.0.downsample.1.bias, head.0.0.0.downsample.1.running_mean, head.0.0.0.downsample.1.running_var, head.0.1.0.conv1.weight, head.0.1.0.bn1.weight, head.0.1.0.bn1.bias, head.0.1.0.bn1.running_mean, head.0.1.0.bn1.running_var, head.0.1.0.conv2.weight, head.0.1.0.bn2.weight, head.0.1.0.bn2.bias, head.0.1.0.bn2.running_mean, head.0.1.0.bn2.running_var, head.0.1.0.conv3.weight, head.0.1.0.bn3.weight, head.0.1.0.bn3.bias, head.0.1.0.bn3.running_mean, head.0.1.0.bn3.running_var, head.0.1.0.downsample.0.weight, head.0.1.0.downsample.1.weight, head.0.1.0.downsample.1.bias, head.0.1.0.downsample.1.running_mean, head.0.1.0.downsample.1.running_var, head.0.2.0.conv1.weight, head.0.2.0.bn1.weight, head.0.2.0.bn1.bias, head.0.2.0.bn1.running_mean, head.0.2.0.bn1.running_var, head.0.2.0.conv2.weight, head.0.2.0.bn2.weight, head.0.2.0.bn2.bias, head.0.2.0.bn2.running_mean, head.0.2.0.bn2.running_var, head.0.2.0.conv3.weight, head.0.2.0.bn3.weight, head.0.2.0.bn3.bias, head.0.2.0.bn3.running_mean, head.0.2.0.bn3.running_var, head.0.2.0.downsample.0.weight, head.0.2.0.downsample.1.weight, head.0.2.0.downsample.1.bias, head.0.2.0.downsample.1.running_mean, head.0.2.0.downsample.1.running_var, head.1.0.0.conv1.weight, head.1.0.0.bn1.weight, head.1.0.0.bn1.bias, head.1.0.0.bn1.running_mean, head.1.0.0.bn1.running_var, head.1.0.0.conv2.weight, head.1.0.0.bn2.weight, head.1.0.0.bn2.bias, head.1.0.0.bn2.running_mean, head.1.0.0.bn2.running_var, head.1.0.0.conv3.weight, head.1.0.0.bn3.weight, head.1.0.0.bn3.bias, head.1.0.0.bn3.running_mean, head.1.0.0.bn3.running_var, head.1.0.0.downsample.0.weight, head.1.0.0.downsample.1.weight, head.1.0.0.downsample.1.bias, head.1.0.0.downsample.1.running_mean, head.1.0.0.downsample.1.running_var, head.1.1.0.conv1.weight, head.1.1.0.bn1.weight, head.1.1.0.bn1.bias, head.1.1.0.bn1.running_mean, head.1.1.0.bn1.running_var, head.1.1.0.conv2.weight, head.1.1.0.bn2.weight, head.1.1.0.bn2.bias, head.1.1.0.bn2.running_mean, head.1.1.0.bn2.running_var, head.1.1.0.conv3.weight, head.1.1.0.bn3.weight, head.1.1.0.bn3.bias, head.1.1.0.bn3.running_mean, head.1.1.0.bn3.running_var, head.1.1.0.downsample.0.weight, head.1.1.0.downsample.1.weight, head.1.1.0.downsample.1.bias, head.1.1.0.downsample.1.running_mean, head.1.1.0.downsample.1.running_var, head.2.0.0.conv1.weight, head.2.0.0.bn1.weight, head.2.0.0.bn1.bias, head.2.0.0.bn1.running_mean, head.2.0.0.bn1.running_var, head.2.0.0.conv2.weight, head.2.0.0.bn2.weight, head.2.0.0.bn2.bias, head.2.0.0.bn2.running_mean, head.2.0.0.bn2.running_var, head.2.0.0.conv3.weight, head.2.0.0.bn3.weight, head.2.0.0.bn3.bias, head.2.0.0.bn3.running_mean, head.2.0.0.bn3.running_var, head.2.0.0.downsample.0.weight, head.2.0.0.downsample.1.weight, head.2.0.0.downsample.1.bias, head.2.0.0.downsample.1.running_mean, head.2.0.0.downsample.1.running_var, head.3.0.0.conv1.weight, head.3.0.0.bn1.weight, head.3.0.0.bn1.bias, head.3.0.0.bn1.running_mean, head.3.0.0.bn1.running_var, head.3.0.0.conv2.weight, head.3.0.0.bn2.weight, head.3.0.0.bn2.bias, head.3.0.0.bn2.running_mean, head.3.0.0.bn2.running_var, head.3.0.0.conv3.weight, head.3.0.0.bn3.weight, head.3.0.0.bn3.bias, head.3.0.0.bn3.running_mean, head.3.0.0.bn3.running_var, head.3.0.0.downsample.0.weight, head.3.0.0.downsample.1.weight, head.3.0.0.downsample.1.bias, head.3.0.0.downsample.1.running_mean, head.3.0.0.downsample.1.running_var, fc.weight, fc.bias

The maximum bach size I can do with 1024 image size is 3 Sylvain Ard 054950772 4 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:32, Yining Li @.***> a écrit :

BTW, I would suggest starting from a simpler model like a top-down heatmap model instead of CID. The former should be easier to tune on your own data. Maybe HRNet https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md is what you may be interested in. Of course, you would need to train a detection model to extract object bbox for a top-down keypoint model.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333478955, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR4B3EALV7A6GPNXNILWLBWDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

after 140 epochs with batch size =25 the results are :

2022-12-01 13:13:55,455 - mmpose - INFO - Epoch [140][50/51] lr: 1.000e-05, eta: 0:00:00, time: 0.801, data_time: 0.208, memory: 13346, multi_heatmap_loss: 0.4749, single_heatmap_loss: 1.9669, contrastive_loss: 0.0246, loss: 2.4665 2022-12-01 13:13:56,065 - mmpose - INFO - Saving checkpoint at 140 epochs [ ] 0/101, elapsed: 0s, ETA:C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv__init.py:20: UserWarning: On January 1, 2023, MMCV will release v2.0.0, in which it will remove components related to the training process and add a data transformation module. In addition, it will rename the package names mmcv to mmcv-lite and mmcv-full to mmcv. See https://github.com/open-mmlab/mmcv/blob/master/docs/en/compatibility.md for more details. warnings.warn( C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv\init__.py:20: UserWarning: On January 1, 2023, MMCV will release v2.0.0, in which it will remove components related to the training process and add a data transformation module. In addition, it will rename the package names mmcv to mmcv-lite and mmcv-full to mmcv. See https://github.com/open-mmlab/mmcv/blob/master/docs/en/compatibility.md for more details. warnings.warn( [>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 101/101, 2.9 task/s, elapsed: 35s, ETA: 0sLoading and preparing results... DONE (t=0.01s) creating index... index created! Running per image evaluation... Evaluate annotation type keypoints DONE (t=0.49s). Accumulating evaluation results... DONE (t=0.01s). Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.357 Average Precision (AP) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.734 Average Precision (AP) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.315 Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.002 Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.373 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.470 Average Recall (AR) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.792 Average Recall (AR) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.478 Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.100 Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.472 2022-12-01 13:14:35,225 - mmpose - INFO - The previous best checkpoint C:\mmpose-master\output\best_AP_epoch_120.pth was removed 2022-12-01 13:14:37,579 - mmpose - INFO - Now best checkpoint is saved as best_AP_epoch_140.pth. 2022-12-01 13:14:37,580 - mmpose - INFO - Best AP is 0.3571 at 140 epoch. 2022-12-01 13:14:37,582 - mmpose - INFO - Epoch(val) [140][101] AP: 0.3571, AP .5: 0.7335, AP .75: 0.3148, AP (M): 0.0024, AP (L): 0.3734, AR: 0.4702, AR .5: 0.7923, AR .75: 0.4777, AR (M): 0.1000, AR (L): 0.4725

I launch it on 450 epochs much Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 12:30, Sylvain Ard @.***> a écrit :

and if we change the learning rate ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 11:23, Sylvain Ard @.***> a écrit :

I think the only part where we can do something is :

backbone=dict( type='HRNet', in_channels=3, extra=dict( stage1=dict( num_modules=1, num_branches=1, block='BOTTLENECK', num_blocks=(4, ), num_channels=(64, )), stage2=dict( num_modules=1, num_branches=2, block='BASIC', num_blocks=(4, 4), num_channels=(48, 96)), stage3=dict( num_modules=4, num_branches=3, block='BASIC', num_blocks=(4, 4, 4), num_channels=(48, 96, 192)), stage4=dict( num_modules=3, num_branches=4, block='BASIC', num_blocks=(4, 4, 4, 4), num_channels=(48, 96, 192, 384), multiscale_output=True)), ),

but I don't know what, an idea please ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 11:04, Sylvain Ard @.***> a écrit :

The training and testing dataset uses the same annotation file. Is it intended to use the same data for both training and testing? >> yes

The training batch size is set as 1 (assume 1 GPU is used) which may be too small. >> I tried 3 but I don't see the progression, and I don't know if it advance or not

There are lots of log files in the output folder and I am not sure which one to check.

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:57, Sylvain Ard @.***> a écrit :

when I do batch_size=3 the training does not advance Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:48, Sylvain Ard @.***> a écrit :

I don't think the model you suggest me is simpler than cid, the parameters are the same Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:45, Sylvain Ard @.***> a écrit :

the logs are very similar there are two warnings but I don't understand these :

2022-11-25 14:48:28,267 - mmpose - WARNING - The model and loaded state dict do not match exactly

unexpected key in source state_dict: head.0.0.0.conv1.weight, head.0.0.0.bn1.weight, head.0.0.0.bn1.bias, head.0.0.0.bn1.running_mean, head.0.0.0.bn1.running_var, head.0.0.0.conv2.weight, head.0.0.0.bn2.weight, head.0.0.0.bn2.bias, head.0.0.0.bn2.running_mean, head.0.0.0.bn2.running_var, head.0.0.0.conv3.weight, head.0.0.0.bn3.weight, head.0.0.0.bn3.bias, head.0.0.0.bn3.running_mean, head.0.0.0.bn3.running_var, head.0.0.0.downsample.0.weight, head.0.0.0.downsample.1.weight, head.0.0.0.downsample.1.bias, head.0.0.0.downsample.1.running_mean, head.0.0.0.downsample.1.running_var, head.0.1.0.conv1.weight, head.0.1.0.bn1.weight, head.0.1.0.bn1.bias, head.0.1.0.bn1.running_mean, head.0.1.0.bn1.running_var, head.0.1.0.conv2.weight, head.0.1.0.bn2.weight, head.0.1.0.bn2.bias, head.0.1.0.bn2.running_mean, head.0.1.0.bn2.running_var, head.0.1.0.conv3.weight, head.0.1.0.bn3.weight, head.0.1.0.bn3.bias, head.0.1.0.bn3.running_mean, head.0.1.0.bn3.running_var, head.0.1.0.downsample.0.weight, head.0.1.0.downsample.1.weight, head.0.1.0.downsample.1.bias, head.0.1.0.downsample.1.running_mean, head.0.1.0.downsample.1.running_var, head.0.2.0.conv1.weight, head.0.2.0.bn1.weight, head.0.2.0.bn1.bias, head.0.2.0.bn1.running_mean, head.0.2.0.bn1.running_var, head.0.2.0.conv2.weight, head.0.2.0.bn2.weight, head.0.2.0.bn2.bias, head.0.2.0.bn2.running_mean, head.0.2.0.bn2.running_var, head.0.2.0.conv3.weight, head.0.2.0.bn3.weight, head.0.2.0.bn3.bias, head.0.2.0.bn3.running_mean, head.0.2.0.bn3.running_var, head.0.2.0.downsample.0.weight, head.0.2.0.downsample.1.weight, head.0.2.0.downsample.1.bias, head.0.2.0.downsample.1.running_mean, head.0.2.0.downsample.1.running_var, head.1.0.0.conv1.weight, head.1.0.0.bn1.weight, head.1.0.0.bn1.bias, head.1.0.0.bn1.running_mean, head.1.0.0.bn1.running_var, head.1.0.0.conv2.weight, head.1.0.0.bn2.weight, head.1.0.0.bn2.bias, head.1.0.0.bn2.running_mean, head.1.0.0.bn2.running_var, head.1.0.0.conv3.weight, head.1.0.0.bn3.weight, head.1.0.0.bn3.bias, head.1.0.0.bn3.running_mean, head.1.0.0.bn3.running_var, head.1.0.0.downsample.0.weight, head.1.0.0.downsample.1.weight, head.1.0.0.downsample.1.bias, head.1.0.0.downsample.1.running_mean, head.1.0.0.downsample.1.running_var, head.1.1.0.conv1.weight, head.1.1.0.bn1.weight, head.1.1.0.bn1.bias, head.1.1.0.bn1.running_mean, head.1.1.0.bn1.running_var, head.1.1.0.conv2.weight, head.1.1.0.bn2.weight, head.1.1.0.bn2.bias, head.1.1.0.bn2.running_mean, head.1.1.0.bn2.running_var, head.1.1.0.conv3.weight, head.1.1.0.bn3.weight, head.1.1.0.bn3.bias, head.1.1.0.bn3.running_mean, head.1.1.0.bn3.running_var, head.1.1.0.downsample.0.weight, head.1.1.0.downsample.1.weight, head.1.1.0.downsample.1.bias, head.1.1.0.downsample.1.running_mean, head.1.1.0.downsample.1.running_var, head.2.0.0.conv1.weight, head.2.0.0.bn1.weight, head.2.0.0.bn1.bias, head.2.0.0.bn1.running_mean, head.2.0.0.bn1.running_var, head.2.0.0.conv2.weight, head.2.0.0.bn2.weight, head.2.0.0.bn2.bias, head.2.0.0.bn2.running_mean, head.2.0.0.bn2.running_var, head.2.0.0.conv3.weight, head.2.0.0.bn3.weight, head.2.0.0.bn3.bias, head.2.0.0.bn3.running_mean, head.2.0.0.bn3.running_var, head.2.0.0.downsample.0.weight, head.2.0.0.downsample.1.weight, head.2.0.0.downsample.1.bias, head.2.0.0.downsample.1.running_mean, head.2.0.0.downsample.1.running_var, head.3.0.0.conv1.weight, head.3.0.0.bn1.weight, head.3.0.0.bn1.bias, head.3.0.0.bn1.running_mean, head.3.0.0.bn1.running_var, head.3.0.0.conv2.weight, head.3.0.0.bn2.weight, head.3.0.0.bn2.bias, head.3.0.0.bn2.running_mean, head.3.0.0.bn2.running_var, head.3.0.0.conv3.weight, head.3.0.0.bn3.weight, head.3.0.0.bn3.bias, head.3.0.0.bn3.running_mean, head.3.0.0.bn3.running_var, head.3.0.0.downsample.0.weight, head.3.0.0.downsample.1.weight, head.3.0.0.downsample.1.bias, head.3.0.0.downsample.1.running_mean, head.3.0.0.downsample.1.running_var, fc.weight, fc.bias

The maximum bach size I can do with 1024 image size is 3 Sylvain Ard 054950772 4 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:32, Yining Li @.***> a écrit :

BTW, I would suggest starting from a simpler model like a top-down heatmap model instead of CID. The former should be easier to tune on your own data. Maybe HRNet https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md is what you may be interested in. Of course, you would need to train a detection model to extract object bbox for a top-down keypoint model.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333478955, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR4B3EALV7A6GPNXNILWLBWDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

batch size=2 perdon Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 13:16, Sylvain Ard @.***> a écrit :

after 140 epochs with batch size =25 the results are :

2022-12-01 13:13:55,455 - mmpose - INFO - Epoch [140][50/51] lr: 1.000e-05, eta: 0:00:00, time: 0.801, data_time: 0.208, memory: 13346, multi_heatmap_loss: 0.4749, single_heatmap_loss: 1.9669, contrastive_loss: 0.0246, loss: 2.4665 2022-12-01 13:13:56,065 - mmpose - INFO - Saving checkpoint at 140 epochs [ ] 0/101, elapsed: 0s, ETA:C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv__init.py:20: UserWarning: On January 1, 2023, MMCV will release v2.0.0, in which it will remove components related to the training process and add a data transformation module. In addition, it will rename the package names mmcv to mmcv-lite and mmcv-full to mmcv. See https://github.com/open-mmlab/mmcv/blob/master/docs/en/compatibility.md for more details. warnings.warn( C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv\init__.py:20: UserWarning: On January 1, 2023, MMCV will release v2.0.0, in which it will remove components related to the training process and add a data transformation module. In addition, it will rename the package names mmcv to mmcv-lite and mmcv-full to mmcv. See https://github.com/open-mmlab/mmcv/blob/master/docs/en/compatibility.md for more details. warnings.warn( [>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 101/101, 2.9 task/s, elapsed: 35s, ETA: 0sLoading and preparing results... DONE (t=0.01s) creating index... index created! Running per image evaluation... Evaluate annotation type keypoints DONE (t=0.49s). Accumulating evaluation results... DONE (t=0.01s). Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.357 Average Precision (AP) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.734 Average Precision (AP) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.315 Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.002 Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.373 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.470 Average Recall (AR) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.792 Average Recall (AR) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.478 Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.100 Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.472 2022-12-01 13:14:35,225 - mmpose - INFO - The previous best checkpoint C:\mmpose-master\output\best_AP_epoch_120.pth was removed 2022-12-01 13:14:37,579 - mmpose - INFO - Now best checkpoint is saved as best_AP_epoch_140.pth. 2022-12-01 13:14:37,580 - mmpose - INFO - Best AP is 0.3571 at 140 epoch. 2022-12-01 13:14:37,582 - mmpose - INFO - Epoch(val) [140][101] AP: 0.3571, AP .5: 0.7335, AP .75: 0.3148, AP (M): 0.0024, AP (L): 0.3734, AR: 0.4702, AR .5: 0.7923, AR .75: 0.4777, AR (M): 0.1000, AR (L): 0.4725

I launch it on 450 epochs much Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 12:30, Sylvain Ard @.***> a écrit :

and if we change the learning rate ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 11:23, Sylvain Ard @.***> a écrit :

I think the only part where we can do something is :

backbone=dict( type='HRNet', in_channels=3, extra=dict( stage1=dict( num_modules=1, num_branches=1, block='BOTTLENECK', num_blocks=(4, ), num_channels=(64, )), stage2=dict( num_modules=1, num_branches=2, block='BASIC', num_blocks=(4, 4), num_channels=(48, 96)), stage3=dict( num_modules=4, num_branches=3, block='BASIC', num_blocks=(4, 4, 4), num_channels=(48, 96, 192)), stage4=dict( num_modules=3, num_branches=4, block='BASIC', num_blocks=(4, 4, 4, 4), num_channels=(48, 96, 192, 384), multiscale_output=True)), ),

but I don't know what, an idea please ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 11:04, Sylvain Ard @.***> a écrit :

The training and testing dataset uses the same annotation file. Is it intended to use the same data for both training and testing? >> yes

The training batch size is set as 1 (assume 1 GPU is used) which may be too small. >> I tried 3 but I don't see the progression, and I don't know if it advance or not

There are lots of log files in the output folder and I am not sure which one to check.

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:57, Sylvain Ard @.***> a écrit :

when I do batch_size=3 the training does not advance Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:48, Sylvain Ard @.***> a écrit :

I don't think the model you suggest me is simpler than cid, the parameters are the same Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:45, Sylvain Ard @.***> a écrit :

the logs are very similar there are two warnings but I don't understand these :

2022-11-25 14:48:28,267 - mmpose - WARNING - The model and loaded state dict do not match exactly

unexpected key in source state_dict: head.0.0.0.conv1.weight, head.0.0.0.bn1.weight, head.0.0.0.bn1.bias, head.0.0.0.bn1.running_mean, head.0.0.0.bn1.running_var, head.0.0.0.conv2.weight, head.0.0.0.bn2.weight, head.0.0.0.bn2.bias, head.0.0.0.bn2.running_mean, head.0.0.0.bn2.running_var, head.0.0.0.conv3.weight, head.0.0.0.bn3.weight, head.0.0.0.bn3.bias, head.0.0.0.bn3.running_mean, head.0.0.0.bn3.running_var, head.0.0.0.downsample.0.weight, head.0.0.0.downsample.1.weight, head.0.0.0.downsample.1.bias, head.0.0.0.downsample.1.running_mean, head.0.0.0.downsample.1.running_var, head.0.1.0.conv1.weight, head.0.1.0.bn1.weight, head.0.1.0.bn1.bias, head.0.1.0.bn1.running_mean, head.0.1.0.bn1.running_var, head.0.1.0.conv2.weight, head.0.1.0.bn2.weight, head.0.1.0.bn2.bias, head.0.1.0.bn2.running_mean, head.0.1.0.bn2.running_var, head.0.1.0.conv3.weight, head.0.1.0.bn3.weight, head.0.1.0.bn3.bias, head.0.1.0.bn3.running_mean, head.0.1.0.bn3.running_var, head.0.1.0.downsample.0.weight, head.0.1.0.downsample.1.weight, head.0.1.0.downsample.1.bias, head.0.1.0.downsample.1.running_mean, head.0.1.0.downsample.1.running_var, head.0.2.0.conv1.weight, head.0.2.0.bn1.weight, head.0.2.0.bn1.bias, head.0.2.0.bn1.running_mean, head.0.2.0.bn1.running_var, head.0.2.0.conv2.weight, head.0.2.0.bn2.weight, head.0.2.0.bn2.bias, head.0.2.0.bn2.running_mean, head.0.2.0.bn2.running_var, head.0.2.0.conv3.weight, head.0.2.0.bn3.weight, head.0.2.0.bn3.bias, head.0.2.0.bn3.running_mean, head.0.2.0.bn3.running_var, head.0.2.0.downsample.0.weight, head.0.2.0.downsample.1.weight, head.0.2.0.downsample.1.bias, head.0.2.0.downsample.1.running_mean, head.0.2.0.downsample.1.running_var, head.1.0.0.conv1.weight, head.1.0.0.bn1.weight, head.1.0.0.bn1.bias, head.1.0.0.bn1.running_mean, head.1.0.0.bn1.running_var, head.1.0.0.conv2.weight, head.1.0.0.bn2.weight, head.1.0.0.bn2.bias, head.1.0.0.bn2.running_mean, head.1.0.0.bn2.running_var, head.1.0.0.conv3.weight, head.1.0.0.bn3.weight, head.1.0.0.bn3.bias, head.1.0.0.bn3.running_mean, head.1.0.0.bn3.running_var, head.1.0.0.downsample.0.weight, head.1.0.0.downsample.1.weight, head.1.0.0.downsample.1.bias, head.1.0.0.downsample.1.running_mean, head.1.0.0.downsample.1.running_var, head.1.1.0.conv1.weight, head.1.1.0.bn1.weight, head.1.1.0.bn1.bias, head.1.1.0.bn1.running_mean, head.1.1.0.bn1.running_var, head.1.1.0.conv2.weight, head.1.1.0.bn2.weight, head.1.1.0.bn2.bias, head.1.1.0.bn2.running_mean, head.1.1.0.bn2.running_var, head.1.1.0.conv3.weight, head.1.1.0.bn3.weight, head.1.1.0.bn3.bias, head.1.1.0.bn3.running_mean, head.1.1.0.bn3.running_var, head.1.1.0.downsample.0.weight, head.1.1.0.downsample.1.weight, head.1.1.0.downsample.1.bias, head.1.1.0.downsample.1.running_mean, head.1.1.0.downsample.1.running_var, head.2.0.0.conv1.weight, head.2.0.0.bn1.weight, head.2.0.0.bn1.bias, head.2.0.0.bn1.running_mean, head.2.0.0.bn1.running_var, head.2.0.0.conv2.weight, head.2.0.0.bn2.weight, head.2.0.0.bn2.bias, head.2.0.0.bn2.running_mean, head.2.0.0.bn2.running_var, head.2.0.0.conv3.weight, head.2.0.0.bn3.weight, head.2.0.0.bn3.bias, head.2.0.0.bn3.running_mean, head.2.0.0.bn3.running_var, head.2.0.0.downsample.0.weight, head.2.0.0.downsample.1.weight, head.2.0.0.downsample.1.bias, head.2.0.0.downsample.1.running_mean, head.2.0.0.downsample.1.running_var, head.3.0.0.conv1.weight, head.3.0.0.bn1.weight, head.3.0.0.bn1.bias, head.3.0.0.bn1.running_mean, head.3.0.0.bn1.running_var, head.3.0.0.conv2.weight, head.3.0.0.bn2.weight, head.3.0.0.bn2.bias, head.3.0.0.bn2.running_mean, head.3.0.0.bn2.running_var, head.3.0.0.conv3.weight, head.3.0.0.bn3.weight, head.3.0.0.bn3.bias, head.3.0.0.bn3.running_mean, head.3.0.0.bn3.running_var, head.3.0.0.downsample.0.weight, head.3.0.0.downsample.1.weight, head.3.0.0.downsample.1.bias, head.3.0.0.downsample.1.running_mean, head.3.0.0.downsample.1.running_var, fc.weight, fc.bias

The maximum bach size I can do with 1024 image size is 3 Sylvain Ard 054950772 4 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 10:32, Yining Li @.***> a écrit :

BTW, I would suggest starting from a simpler model like a top-down heatmap model instead of CID. The former should be easier to tune on your own data. Maybe HRNet https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md is what you may be interested in. Of course, you would need to train a detection model to extract object bbox for a top-down keypoint model.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333478955, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR4B3EALV7A6GPNXNILWLBWDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

ly015 commented 1 year ago

The CID model we provided on the COCO dataset uses a total batch size of 160 (8 GPU) for training. Training with a very small batch size could be unstable or even non-convergent. Maybe consider reducing the image size or the feature channels to have a larger batch size with limited GPU memory.

The top-down heatmap is a straightforward and widely used keypoint detection method, which is usually easy to train and can achieve promising performance in most cases, based on our experience. By 'simpler' I mean the algorithm itself, not how the config or the parameters look like. It's still recommended to try a top-down heatmap model as a starter.

I am afraid the above is all suggestions I can give now.

SylvainArd commented 1 year ago

Ok have you an example of top down algorithm config file to give me please ?

Envoyé de mon iPhone

Le 1 déc. 2022 à 14:13, Yining Li @.***> a écrit :

The CID model we provided on the COCO dataset uses a total batch size of 160 (8 GPU) for training. Training with a very small batch size could be unstable or even non-convergent. Maybe consider reducing the image size or the feature channels to have a larger batch size with limited GPU memory.

The top-down heatmap is a straightforward and widely used keypoint detection method, which is usually easy to train and can achieve promising performance in most cases, based on our experience. By 'simpler' I mean the algorithm itself, not how the config or the parameters look like. It's still recommended to try a top-down heatmap model as a starter.

I am afraid the above is all suggestions I can give now.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.

ly015 commented 1 year ago

The link to a summary page of top-down heatmap models with HRNet backbone has been given above.

And with all respect, I would have to suggest that it's better not to include too much irrelevant content (e.g. email signatures and history messages) in an issue discussion, which can be a distractor for readers and potential helpers from the community.

SylvainArd commented 1 year ago

OK thank you but I see : image_size=[288, 384], the image size is fix if I put [1024,1024] my images which have not the same heights and widths will be deformed isn't it ?

best regards

Le jeu. 1 déc. 2022 à 14:31, Yining Li @.***> a écrit :

The link https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md to a summary page of top-down heatmap models with HRNet has been given above.

And with all respect, I would have to suggest that it's better not to include too much irrelevant content (e.g. email signatures and history messages) in an issue discussion, which can be a distractor for readers and potential helpers from the community.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333775484, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR6XWAOWN5E2TZPVGJ3WLCSDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

huh if I put image_size=[288, 384], sorry because 1024 is too high because of samples_per_gpu Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 16:12, Sylvain Ard @.***> a écrit :

OK thank you but I see : image_size=[288, 384], the image size is fix if I put [1024,1024] my images which have not the same heights and widths will be deformed isn't it ?

best regards

Le jeu. 1 déc. 2022 à 14:31, Yining Li @.***> a écrit :

The link https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md to a summary page of top-down heatmap models with HRNet has been given above.

And with all respect, I would have to suggest that it's better not to include too much irrelevant content (e.g. email signatures and history messages) in an issue discussion, which can be a distractor for readers and potential helpers from the community.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333775484, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR6XWAOWN5E2TZPVGJ3WLCSDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

sorry but I have the error : Traceback (most recent call last): File "C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv\utils\registry.py", line 69, in build_from_cfg return obj_cls(**args) File "c:\mmpose-master\mmpose\datasets\datasets\top_down\topdown_coco_dataset.py", line 75, in init super().init( File "c:\mmpose-master\mmpose\datasets\datasets\base\kpt_2d_sview_rgb_img_top_down_dataset.py", line 74, in init assert self.ann_info['num_joints'] == dataset_info.keypoint_num AssertionError

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "tools/train.py", line 201, in main() File "tools/train.py", line 176, in main datasets = [build_dataset(cfg.data.train)] File "c:\mmpose-master\mmpose\datasets\builder.py", line 87, in build_dataset dataset = build_from_cfg(cfg, DATASETS, default_args) File "C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv\utils\registry.py", line 72, in build_from_cfg raise type(e)(f'{obj_cls.name}: {e}') AssertionError: TopDownCocoDataset: my config file : https://www.dropbox.com/s/b2u89p5onx7f6it/hrnet_w32_coco_384x288.py?dl=0

my config file : I launched : python tools/train.py configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_w48_coco_384x288.py --work-dir "output" Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 16:13, Sylvain Ard @.***> a écrit :

huh if I put image_size=[288, 384], sorry because 1024 is too high because of samples_per_gpu Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 16:12, Sylvain Ard @.***> a écrit :

OK thank you but I see : image_size=[288, 384], the image size is fix if I put [1024,1024] my images which have not the same heights and widths will be deformed isn't it ?

best regards

Le jeu. 1 déc. 2022 à 14:31, Yining Li @.***> a écrit :

The link https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md to a summary page of top-down heatmap models with HRNet has been given above.

And with all respect, I would have to suggest that it's better not to include too much irrelevant content (e.g. email signatures and history messages) in an issue discussion, which can be a distractor for readers and potential helpers from the community.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333775484, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR6XWAOWN5E2TZPVGJ3WLCSDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

sorry for the informations below, I will be attentive now

Le jeu. 1 déc. 2022 à 16:35, Sylvain Ard @.***> a écrit :

sorry but I have the error : Traceback (most recent call last): File "C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv\utils\registry.py", line 69, in build_from_cfg return obj_cls(**args) File "c:\mmpose-master\mmpose\datasets\datasets\top_down\topdown_coco_dataset.py", line 75, in init super().init( File "c:\mmpose-master\mmpose\datasets\datasets\base\kpt_2d_sview_rgb_img_top_down_dataset.py", line 74, in init assert self.ann_info['num_joints'] == dataset_info.keypoint_num AssertionError

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "tools/train.py", line 201, in main() File "tools/train.py", line 176, in main datasets = [build_dataset(cfg.data.train)] File "c:\mmpose-master\mmpose\datasets\builder.py", line 87, in build_dataset dataset = build_from_cfg(cfg, DATASETS, default_args) File "C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv\utils\registry.py", line 72, in build_from_cfg raise type(e)(f'{obj_cls.name}: {e}') AssertionError: TopDownCocoDataset: my config file : https://www.dropbox.com/s/b2u89p5onx7f6it/hrnet_w32_coco_384x288.py?dl=0

my config file : I launched : python tools/train.py configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_w48_coco_384x288.py --work-dir "output" Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 16:13, Sylvain Ard @.***> a écrit :

huh if I put image_size=[288, 384], sorry because 1024 is too high because of samples_per_gpu Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 16:12, Sylvain Ard @.***> a écrit :

OK thank you but I see : image_size=[288, 384], the image size is fix if I put [1024,1024] my images which have not the same heights and widths will be deformed isn't it ?

best regards

Le jeu. 1 déc. 2022 à 14:31, Yining Li @.***> a écrit :

The link https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md to a summary page of top-down heatmap models with HRNet has been given above.

And with all respect, I would have to suggest that it's better not to include too much irrelevant content (e.g. email signatures and history messages) in an issue discussion, which can be a distractor for readers and potential helpers from the community.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333775484, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR6XWAOWN5E2TZPVGJ3WLCSDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

The problem was I changed a bad config file I had a problem with bbox_file='data/coco/person_detection_results/' 'COCO_val2017_detections_AP_H_56_person.json so I did use_gt_bbox=True, as it described in https://github.com/open-mmlab/mmpose/issues/687

Le jeu. 1 déc. 2022 à 16:36, Sylvain Ard @.***> a écrit :

sorry for the informations below, I will be attentive now

Le jeu. 1 déc. 2022 à 16:35, Sylvain Ard @.***> a écrit :

sorry but I have the error : Traceback (most recent call last): File "C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv\utils\registry.py", line 69, in build_from_cfg return obj_cls(**args) File "c:\mmpose-master\mmpose\datasets\datasets\top_down\topdown_coco_dataset.py", line 75, in init super().init( File "c:\mmpose-master\mmpose\datasets\datasets\base\kpt_2d_sview_rgb_img_top_down_dataset.py", line 74, in init assert self.ann_info['num_joints'] == dataset_info.keypoint_num AssertionError

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "tools/train.py", line 201, in main() File "tools/train.py", line 176, in main datasets = [build_dataset(cfg.data.train)] File "c:\mmpose-master\mmpose\datasets\builder.py", line 87, in build_dataset dataset = build_from_cfg(cfg, DATASETS, default_args) File "C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv\utils\registry.py", line 72, in build_from_cfg raise type(e)(f'{obj_cls.name}: {e}') AssertionError: TopDownCocoDataset: my config file : https://www.dropbox.com/s/b2u89p5onx7f6it/hrnet_w32_coco_384x288.py?dl=0

my config file : I launched : python tools/train.py configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_w48_coco_384x288.py --work-dir "output" Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 16:13, Sylvain Ard @.***> a écrit :

huh if I put image_size=[288, 384], sorry because 1024 is too high because of samples_per_gpu Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 16:12, Sylvain Ard @.***> a écrit :

OK thank you but I see : image_size=[288, 384], the image size is fix if I put [1024,1024] my images which have not the same heights and widths will be deformed isn't it ?

best regards

Le jeu. 1 déc. 2022 à 14:31, Yining Li @.***> a écrit :

The link https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md to a summary page of top-down heatmap models with HRNet has been given above.

And with all respect, I would have to suggest that it's better not to include too much irrelevant content (e.g. email signatures and history messages) in an issue discussion, which can be a distractor for readers and potential helpers from the community.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333775484, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR6XWAOWN5E2TZPVGJ3WLCSDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

here are the results after 20 epochs there are very better :

Le jeu. 1 déc. 2022 à 17:27, Sylvain Ard @.***> a écrit :

The problem was I changed a bad config file I had a problem with bbox_file='data/coco/person_detection_results/' 'COCO_val2017_detections_AP_H_56_person.json so I did use_gt_bbox=True, as it described in https://github.com/open-mmlab/mmpose/issues/687

Le jeu. 1 déc. 2022 à 16:36, Sylvain Ard @.***> a écrit :

sorry for the informations below, I will be attentive now

Le jeu. 1 déc. 2022 à 16:35, Sylvain Ard @.***> a écrit :

sorry but I have the error : Traceback (most recent call last): File "C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv\utils\registry.py", line 69, in build_from_cfg return obj_cls(**args) File "c:\mmpose-master\mmpose\datasets\datasets\top_down\topdown_coco_dataset.py", line 75, in init super().init( File "c:\mmpose-master\mmpose\datasets\datasets\base\kpt_2d_sview_rgb_img_top_down_dataset.py", line 74, in init assert self.ann_info['num_joints'] == dataset_info.keypoint_num AssertionError

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "tools/train.py", line 201, in main() File "tools/train.py", line 176, in main datasets = [build_dataset(cfg.data.train)] File "c:\mmpose-master\mmpose\datasets\builder.py", line 87, in build_dataset dataset = build_from_cfg(cfg, DATASETS, default_args) File "C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv\utils\registry.py", line 72, in build_from_cfg raise type(e)(f'{obj_cls.name}: {e}') AssertionError: TopDownCocoDataset: my config file : https://www.dropbox.com/s/b2u89p5onx7f6it/hrnet_w32_coco_384x288.py?dl=0

my config file : I launched : python tools/train.py configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_w48_coco_384x288.py --work-dir "output" Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 16:13, Sylvain Ard @.***> a écrit :

huh if I put image_size=[288, 384], sorry because 1024 is too high because of samples_per_gpu Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 16:12, Sylvain Ard @.***> a écrit :

OK thank you but I see : image_size=[288, 384], the image size is fix if I put [1024,1024] my images which have not the same heights and widths will be deformed isn't it ?

best regards

Le jeu. 1 déc. 2022 à 14:31, Yining Li @.***> a écrit :

The link https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md to a summary page of top-down heatmap models with HRNet has been given above.

And with all respect, I would have to suggest that it's better not to include too much irrelevant content (e.g. email signatures and history messages) in an issue discussion, which can be a distractor for readers and potential helpers from the community.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333775484, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR6XWAOWN5E2TZPVGJ3WLCSDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

but how to interpret these results please ?

Le jeu. 1 déc. 2022 à 18:08, Sylvain Ard @.***> a écrit :

here are the results after 20 epochs there are very better :

Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.505 Average Precision (AP) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.845 Average Precision (AP) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.535 Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.010 Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.518 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.589 Average Recall (AR) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.880 Average Recall (AR) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.642 Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.171 Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.592 Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 17:27, Sylvain Ard @.***> a écrit :

The problem was I changed a bad config file I had a problem with bbox_file='data/coco/person_detection_results/' 'COCO_val2017_detections_AP_H_56_person.json so I did use_gt_bbox=True, as it described in https://github.com/open-mmlab/mmpose/issues/687

Le jeu. 1 déc. 2022 à 16:36, Sylvain Ard @.***> a écrit :

sorry for the informations below, I will be attentive now

Le jeu. 1 déc. 2022 à 16:35, Sylvain Ard @.***> a écrit :

sorry but I have the error : Traceback (most recent call last): File "C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv\utils\registry.py", line 69, in build_from_cfg return obj_cls(**args) File "c:\mmpose-master\mmpose\datasets\datasets\top_down\topdown_coco_dataset.py", line 75, in init super().init( File "c:\mmpose-master\mmpose\datasets\datasets\base\kpt_2d_sview_rgb_img_top_down_dataset.py", line 74, in init assert self.ann_info['num_joints'] == dataset_info.keypoint_num AssertionError

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "tools/train.py", line 201, in main() File "tools/train.py", line 176, in main datasets = [build_dataset(cfg.data.train)] File "c:\mmpose-master\mmpose\datasets\builder.py", line 87, in build_dataset dataset = build_from_cfg(cfg, DATASETS, default_args) File "C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv\utils\registry.py", line 72, in build_from_cfg raise type(e)(f'{obj_cls.name}: {e}') AssertionError: TopDownCocoDataset: my config file : https://www.dropbox.com/s/b2u89p5onx7f6it/hrnet_w32_coco_384x288.py?dl=0

my config file : I launched : python tools/train.py configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_w48_coco_384x288.py --work-dir "output" Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 16:13, Sylvain Ard @.***> a écrit :

huh if I put image_size=[288, 384], sorry because 1024 is too high because of samples_per_gpu Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 16:12, Sylvain Ard @.***> a écrit :

OK thank you but I see : image_size=[288, 384], the image size is fix if I put [1024,1024] my images which have not the same heights and widths will be deformed isn't it ?

best regards

Le jeu. 1 déc. 2022 à 14:31, Yining Li @.***> a écrit :

The link https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md to a summary page of top-down heatmap models with HRNet has been given above.

And with all respect, I would have to suggest that it's better not to include too much irrelevant content (e.g. email signatures and history messages) in an issue discussion, which can be a distractor for readers and potential helpers from the community.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333775484, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR6XWAOWN5E2TZPVGJ3WLCSDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

hello my results at the end of training are : 2022-12-01 23:00:56,950 - mmpose - INFO - Best AP is 0.9750 at 210 epoch. 2022-12-01 23:00:56,951 - mmpose - INFO - Epoch(val) [210][36] AP: 0.9750, AP .5: 1.0000, AP .75: 0.9901, AP (M): 0.8867, AP (L): 0.9764, AR: 0.9859, AR .5: 1.0000, AR .75: 0.9991, AR (M): 0.9571, AR (L): 0.9861

I have a question : I want to see the results so I did : python demo/top_down_img_demo.py configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_w48_coco_384x288.py "output/latest.pth" --img-root dataset/photos --json-file dataset/coco.json --out-img-root res_img

the results seems perfect but are they ground truth (in this case why give the model at the command) or predicted (in this case why give the JSON of ground truth) ?

I have also run this command :

python tools/test.py configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_w48_coco_384x288.py "output/latest.pth" --out res_img/res.json

I can't see the results on images but I opened the image with Gimp and pointed the cursor on predicted coordinates in JSON and it seems good

Thank you very much

If the results of python tools/test.py configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_w48_coco_384x288.py "output/latest.pth" --out res_img/res.json are predictions your program is perfect !

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 18:58, Sylvain Ard @.***> a écrit :

but how to interpret these results please ?

Le jeu. 1 déc. 2022 à 18:08, Sylvain Ard @.***> a écrit :

here are the results after 20 epochs there are very better :

Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.505 Average Precision (AP) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.845 Average Precision (AP) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.535 Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.010 Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.518 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.589 Average Recall (AR) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.880 Average Recall (AR) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.642 Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.171 Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.592 Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 17:27, Sylvain Ard @.***> a écrit :

The problem was I changed a bad config file I had a problem with bbox_file='data/coco/person_detection_results/' 'COCO_val2017_detections_AP_H_56_person.json so I did use_gt_bbox=True, as it described in https://github.com/open-mmlab/mmpose/issues/687

Le jeu. 1 déc. 2022 à 16:36, Sylvain Ard @.***> a écrit :

sorry for the informations below, I will be attentive now

Le jeu. 1 déc. 2022 à 16:35, Sylvain Ard @.***> a écrit :

sorry but I have the error : Traceback (most recent call last): File "C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv\utils\registry.py", line 69, in build_from_cfg return obj_cls(**args) File "c:\mmpose-master\mmpose\datasets\datasets\top_down\topdown_coco_dataset.py", line 75, in init super().init( File "c:\mmpose-master\mmpose\datasets\datasets\base\kpt_2d_sview_rgb_img_top_down_dataset.py", line 74, in init assert self.ann_info['num_joints'] == dataset_info.keypoint_num AssertionError

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "tools/train.py", line 201, in main() File "tools/train.py", line 176, in main datasets = [build_dataset(cfg.data.train)] File "c:\mmpose-master\mmpose\datasets\builder.py", line 87, in build_dataset dataset = build_from_cfg(cfg, DATASETS, default_args) File "C:\Users\MASTER.conda\envs\openmmlab\lib\site-packages\mmcv\utils\registry.py", line 72, in build_from_cfg raise type(e)(f'{obj_cls.name}: {e}') AssertionError: TopDownCocoDataset: my config file :

https://www.dropbox.com/s/b2u89p5onx7f6it/hrnet_w32_coco_384x288.py?dl=0

my config file : I launched : python tools/train.py configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_w48_coco_384x288.py --work-dir "output" Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 16:13, Sylvain Ard @.***> a écrit :

huh if I put image_size=[288, 384], sorry because 1024 is too high because of samples_per_gpu Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 16:12, Sylvain Ard @.***> a écrit :

OK thank you but I see : image_size=[288, 384], the image size is fix if I put [1024,1024] my images which have not the same heights and widths will be deformed isn't it ?

best regards

Le jeu. 1 déc. 2022 à 14:31, Yining Li @.***> a écrit :

The link https://github.com/open-mmlab/mmpose/blob/master/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_coco.md to a summary page of top-down heatmap models with HRNet has been given above.

And with all respect, I would have to suggest that it's better not to include too much irrelevant content (e.g. email signatures and history messages) in an issue discussion, which can be a distractor for readers and potential helpers from the community.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1333775484, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR6XWAOWN5E2TZPVGJ3WLCSDJANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

ly015 commented 1 year ago

You needn't worry about the ground-truth json file passed to the demo script demo/top_down_img_demo.py. As you may already know, a top-down keypoint model needs the input of the object bounding box to perform the keypoint detection within the bbox area. The ground-truth json file is only used to give the bbox information and the keypoints in the visualized images are predicted by the model. And that's why you would need a detection model together with the top-down keypoint model to perform bbox+keypoint prediction on raw images (where ground-truth boxes are not available).

In the config of a top-down keypoint model in MMPose, the image_size is actually the model input size to which the bbox area will be resized (by affine transform), and it's irrelevant to the original image size. The original image can be in arbitrary sizes.

SylvainArd commented 1 year ago

so for prediction we need to detect the bbox first, the program don't do it ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 08:14, Yining Li @.***> a écrit :

You needn't worry about the ground-truth json file passed to the demo script demo/top_down_img_demo.py. As you may already know that a top-down keypoint model needs the input of the object bounding box to perform the keypoint detection within the bbox area. The ground-truth json file is only used to give the bbox information and the keypoints in the visualized images are predicted by the model. And that's why you would need a detection model together with the top-down keypoint model to perform bbox+keypoint prediction on raw images (where ground-truth boxes are not available).

In the config of a top-down keypoint model in MMPose, the image_size is actually the model input size to which the bbox area will be resized (by affine transform), and it's irrelevant to the original image size. The original image can be in arbitrary sizes.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1334846221, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCRZJTAUKMY5PA2NMIV3WLGOUHANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

I tried to launch test on a new image but it crashed because it doesn't find the image in the JSON, have I to create a JSON with bbox (and obviously not keypoints) for the testing with another program ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 10:36, Sylvain Ard @.***> a écrit :

so for prediction we need to detect the bbox first, the program don't do it ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 08:14, Yining Li @.***> a écrit :

You needn't worry about the ground-truth json file passed to the demo script demo/top_down_img_demo.py. As you may already know that a top-down keypoint model needs the input of the object bounding box to perform the keypoint detection within the bbox area. The ground-truth json file is only used to give the bbox information and the keypoints in the visualized images are predicted by the model. And that's why you would need a detection model together with the top-down keypoint model to perform bbox+keypoint prediction on raw images (where ground-truth boxes are not available).

In the config of a top-down keypoint model in MMPose, the image_size is actually the model input size to which the bbox area will be resized (by affine transform), and it's irrelevant to the original image size. The original image can be in arbitrary sizes.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1334846221, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCRZJTAUKMY5PA2NMIV3WLGOUHANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

well, I have maked a JSON file with bbox but with testing I have the error : KeyError: 'keypoints' it's cheating if we pass the points to the program it's normal that it discovers them! this is not normal at all! Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 10:50, Sylvain Ard @.***> a écrit :

I tried to launch test on a new image but it crashed because it doesn't find the image in the JSON, have I to create a JSON with bbox (and obviously not keypoints) for the testing with another program ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 10:36, Sylvain Ard @.***> a écrit :

so for prediction we need to detect the bbox first, the program don't do it ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 08:14, Yining Li @.***> a écrit :

You needn't worry about the ground-truth json file passed to the demo script demo/top_down_img_demo.py. As you may already know that a top-down keypoint model needs the input of the object bounding box to perform the keypoint detection within the bbox area. The ground-truth json file is only used to give the bbox information and the keypoints in the visualized images are predicted by the model. And that's why you would need a detection model together with the top-down keypoint model to perform bbox+keypoint prediction on raw images (where ground-truth boxes are not available).

In the config of a top-down keypoint model in MMPose, the image_size is actually the model input size to which the bbox area will be resized (by affine transform), and it's irrelevant to the original image size. The original image can be in arbitrary sizes.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1334846221, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCRZJTAUKMY5PA2NMIV3WLGOUHANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

please help me ! it is the final step

Le ven. 2 déc. 2022 à 11:00, Sylvain Ard @.***> a écrit :

well, I have maked a JSON file with bbox but with testing I have the error : KeyError: 'keypoints' it's cheating if we pass the points to the program it's normal that it discovers them! this is not normal at all! Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 10:50, Sylvain Ard @.***> a écrit :

I tried to launch test on a new image but it crashed because it doesn't find the image in the JSON, have I to create a JSON with bbox (and obviously not keypoints) for the testing with another program ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 10:36, Sylvain Ard @.***> a écrit :

so for prediction we need to detect the bbox first, the program don't do it ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 08:14, Yining Li @.***> a écrit :

You needn't worry about the ground-truth json file passed to the demo script demo/top_down_img_demo.py. As you may already know that a top-down keypoint model needs the input of the object bounding box to perform the keypoint detection within the bbox area. The ground-truth json file is only used to give the bbox information and the keypoints in the visualized images are predicted by the model. And that's why you would need a detection model together with the top-down keypoint model to perform bbox+keypoint prediction on raw images (where ground-truth boxes are not available).

In the config of a top-down keypoint model in MMPose, the image_size is actually the model input size to which the bbox area will be resized (by affine transform), and it's irrelevant to the original image size. The original image can be in arbitrary sizes.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1334846221, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCRZJTAUKMY5PA2NMIV3WLGOUHANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

ly015 commented 1 year ago

Yes, top-down keypoint detection needs an object detector. Please see /demo/top_down_img_demo_with_mmdet.py for this paradigm. In this demo, we use MMDetection for detection.

In /demo/top_down_image_demo.py, as I mentioned, only bbox information from the json file is used. You can check it in the code: https://github.com/open-mmlab/mmpose/blob/master/demo/top_down_img_demo.py#L91-L96

If you are using the testing tool, it requires the ground-truth keypoint coordinates to evaluate the model performance, like calculating COCO AP.

SylvainArd commented 1 year ago

OK so how to do a prediction with only bboxes if test needs ground-truth ?

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 11:26, Yining Li @.***> a écrit :

Yes, top-down keypoint detection needs an object detector. Please see /demo/top_down_img_demo_with_mmdet.py for this paradigm. In this demo, we use MMDetection for detection.

In /demo/top_down_image_demo.py, as I mentioned, only bbox information from the json file is used. You can check it in the code: https://github.com/open-mmlab/mmpose/blob/master/demo/top_down_img_demo.py#L91-L96

If you are using the testing tool, it requires the ground-truth keypoint coordinates to evaluate the model performance, like calculating COCO AP.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1335043842, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR7ECOO342V2G6RVIFLWLHFGXANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

what is the command ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 11:30, Sylvain Ard @.***> a écrit :

OK so how to do a prediction with only bboxes if test needs ground-truth ?

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 11:26, Yining Li @.***> a écrit :

Yes, top-down keypoint detection needs an object detector. Please see /demo/top_down_img_demo_with_mmdet.py for this paradigm. In this demo, we use MMDetection for detection.

In /demo/top_down_image_demo.py, as I mentioned, only bbox information from the json file is used. You can check it in the code: https://github.com/open-mmlab/mmpose/blob/master/demo/top_down_img_demo.py#L91-L96

If you are using the testing tool, it requires the ground-truth keypoint coordinates to evaluate the model performance, like calculating COCO AP.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1335043842, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR7ECOO342V2G6RVIFLWLHFGXANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

I mean a prediction with a JSON file ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 11:31, Sylvain Ard @.***> a écrit :

what is the command ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 11:30, Sylvain Ard @.***> a écrit :

OK so how to do a prediction with only bboxes if test needs ground-truth ?

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 11:26, Yining Li @.***> a écrit :

Yes, top-down keypoint detection needs an object detector. Please see /demo/top_down_img_demo_with_mmdet.py for this paradigm. In this demo, we use MMDetection for detection.

In /demo/top_down_image_demo.py, as I mentioned, only bbox information from the json file is used. You can check it in the code: https://github.com/open-mmlab/mmpose/blob/master/demo/top_down_img_demo.py#L91-L96

If you are using the testing tool, it requires the ground-truth keypoint coordinates to evaluate the model performance, like calculating COCO AP.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1335043842, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR7ECOO342V2G6RVIFLWLHFGXANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

ly015 commented 1 year ago

I am confused about your need.

If you want to predict keypoints from an image with a top-down keypoint model, only bounding boxes are needed, no matter if they are from the json file or an extra detector. This is shown in demo/top_down_image_demo.py and demo/top_down_image_demo_with_mmdet.py. And a brief guide is provided in demo/docs/2d_human_pose_demo.md.

If you want to evaluate your model, you need to compare the model prediction with the ground-truth keypoint annotations. That's what the tools/test.sh does.

SylvainArd commented 1 year ago

yes but demo/top_down_image_demo.py only give images, what I want is a JSON where I can find the points coordinates Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 11:38, Yining Li @.***> a écrit :

I am confused about your need.

If you want to predict keypoints from an image with a top-down keypoint model, only bounding boxes are needed, no matter if they are from the json file or an extra detector. This is shown in demo/top_down_image_demo.py and demo/top_down_image_demo_with_mmdet.py. And a brief guide is provided in demo/docs/2d_human_pose_demo.md.

If you want to evaluate your model, you need to compare the model prediction with the ground-truth keypoint annotations. That's what the tools/test.sh does.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1335055777, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCRYQ4HONDRFZ2TACDNDWLHGTHANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

ly015 commented 1 year ago

Maybe just add a few lines of code to dump the prediction to a JSON? The demo is just meant to show the basic use of MMPose API.

SylvainArd commented 1 year ago

on demo on an image of training dataset wich I give only bboxes it detects none keypoints, whereas when I give keypoints it detects the keypoints and display it !! Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 11:43, Yining Li @.***> a écrit :

Maybe just add a few lines of code to dump the prediction to a JSON? The demo is just meant to show the basic use of MMPose API.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1335061212, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR7JT5QJIPRRLJ2EKUTWLHHF3ANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

it works ! it was errors in my JSON file

Le ven. 2 déc. 2022 à 12:08, Sylvain Ard @.***> a écrit :

on demo on an image of training dataset wich I give only bboxes it detects none keypoints, whereas when I give keypoints it detects the keypoints and display it !! Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 11:43, Yining Li @.***> a écrit :

Maybe just add a few lines of code to dump the prediction to a JSON? The demo is just meant to show the basic use of MMPose API.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1335061212, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR7JT5QJIPRRLJ2EKUTWLHHF3ANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

SylvainArd commented 1 year ago

I added the lines :

    chaine=""
    for line in pose_results:
        bbox=line["bbox"]
        kps=line["keypoints"]
        chaine=chaine+"bbox:"
        i=0
        for c in bbox:
            i=i+1
            chaine=chaine+str(c)
            if i!=4:
                chaine=chaine+","
        chaine=chaine+"|keypoints:["
        for kp in kps:
            chaine=chaine+"["
            i=0
            for c in kp:
                i=i+1
                chaine=chaine+str(c)
                if i!=3:
                    chaine=chaine+","
            chaine=chaine+"]"
        chaine=chaine+"]\n"

    with open('demo_img/kp.txt', 'w') as f:
        f.write(chaine)

to top_down_img_demo.py it works, you can close the ticket, a last question the image size is 384x288, if the input image is more high than wide, it is deformed isn'it and it decrease training efficiency isn'it ?

Le ven. 2 déc. 2022 à 12:12, Sylvain Ard @.***> a écrit :

it works ! it was errors in my JSON file

Le ven. 2 déc. 2022 à 12:08, Sylvain Ard @.***> a écrit :

on demo on an image of training dataset wich I give only bboxes it detects none keypoints, whereas when I give keypoints it detects the keypoints and display it !! Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 2 déc. 2022 à 11:43, Yining Li @.***> a écrit :

Maybe just add a few lines of code to dump the prediction to a JSON? The demo is just meant to show the basic use of MMPose API.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1841#issuecomment-1335061212, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR7JT5QJIPRRLJ2EKUTWLHHF3ANCNFSM6AAAAAASLKEXGM . You are receiving this because you authored the thread.Message ID: @.***>

ly015 commented 1 year ago

As explained above, 384x288 is the model input size to which the bbox area will be resized, and it's irrelevant to the full image size. And the size 384x288 is for human keypoint detection (that's why the height is larger than the width). You can set it according to the task.