missing model weight when training on visualgenome

[entalent]

I attempted to train RelDN model on visual genome by executing this command:

python tools/train_sg_net.py --config-file sgg_configs/vg_vrd/rel_danfeiX_FPN50_reldn.yaml

and get the following error:

FileNotFoundError: [Errno 2] No such file or directory: 'models/vgvrd/vgnm_usefpTrue_objctx0_edgectx2/model_final.pth'

I searched in the repository but found no information about the file vgnm_usefpTrue_objctx0_edgectx2/model_final.pth. Could you please provide more details about where can we download this model file? Thanks in advance.

[entalent]

In the training log of the RelDN model pre-trained on visual genome (sgg_benchmark/sgg_model_zoo/visualgenome/reldn_usefpTrue_lr0.005_bsz4/log.txt, line 42), it seems that the training process on visual genome starts from a checkpoint pretrained_model/multitask_pengchuan/frcnn_x152fpn_4sets.yaml_pos0.5_lr0.005_bsz16.pth.

I want to train the RelDN model from scratch on another dataset, and I would be more than grateful if you could share the checkpoint where the training starts.

[tao123322]

Do you have the weighting document? Can you send me a copy?

[hanxiaotian]

The model path in side the config yaml are already pretrained and can be used to run directly. If you want to train from scratch, you can set the model path to be pretrained maskrcnn-benchmark object detection model's weight

[tao123322]

Thanks for the teacher's reply. But I encountered a bigger problem in the process of training reldn. I have put the specific questions in the attachment and presented them as pictures, please help me to take a look. I hope you will reply to me as soon as possible, thank you teacher!

------------------ 原始邮件 ------------------ 发件人: "microsoft/scene_graph_benchmark" @.>; 发送时间: 2021年10月14日(星期四) 中午11:47 @.>; @.**@.>; 主题: Re: [microsoft/scene_graph_benchmark] missing model weight when training on visualgenome (#45)

The model path in side the config yaml are already pretrained and can be used to run directly. If you want to train from scratch, you can set the model path to be pretrained maskrcnn-benchmark object detection model's weight

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

[tao123322]

Hello, teacher, when training reldn, there is an error in line 39 of the code in the maskrcnn_benchmark/engine/inference.py file

Cause a Runtime error, what is the cause of this. The cuda11.0 I use, does it have anything to do with this?

------------------ 原始邮件 ------------------ 发件人: "microsoft/scene_graph_benchmark" @.>; 发送时间: 2021年10月14日(星期四) 中午11:47 @.>; @.**@.>; 主题: Re: [microsoft/scene_graph_benchmark] missing model weight when training on visualgenome (#45)

The model path in side the config yaml are already pretrained and can be used to run directly. If you want to train from scratch, you can set the model path to be pretrained maskrcnn-benchmark object detection model's weight

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

[Ceazy-Gentleman]

@entalent Do you solve the problem? I also want to train a model from scratch with the pretrained object dectection model.

I attempted to train RelDN model on visual genome by executing this command:

python tools/train_sg_net.py --config-file sgg_configs/vg_vrd/rel_danfeiX_FPN50_reldn.yaml

and get the following error:

FileNotFoundError: [Errno 2] No such file or directory: 'models/vgvrd/vgnm_usefpTrue_objctx0_edgectx2/model_final.pth'

I searched in the repository but found no information about the file vgnm_usefpTrue_objctx0_edgectx2/model_final.pth. Could you please provide more details about where can we download this model file? Thanks in advance. Can you tell me how to down load the pretained model?