Nicholasli1995 / EvoSkeleton

Official project website for the CVPR 2020 paper (Oral Presentation) "Cascaded deep monocular 3D human pose estimation wth evolutionary training data"
https://arxiv.org/abs/2006.07778
MIT License
333 stars 44 forks source link

2D Pose Estimation #18

Closed leilaUEA closed 3 years ago

leilaUEA commented 3 years ago

Hi, Is there a possibility to do 2D pose estimation of an arbitrary image with the current code that is uploaded? the inference code is uploading skeleton from a fixed, predefined dictionary of annotation.... Thanks very much in advance Best regards, Leila

Nicholasli1995 commented 3 years ago

Hi, I will upload the pre-trained 2D pose estimator used in the paper tommorrow. However, this pre-trained model is only trained on H36M and may not be perfect for in-the-wild images. To get good performance on outdoor images, you can try:

  1. Use other models trained on in-the-wild images.
  2. Fine-tune my model with annotated outdoor images (like the provided U3DPW dataset).
leilaUEA commented 3 years ago

Thanks very much for your response and for uploading the 2D pose estimator, I appreciate it!

I also tried running the annotator, annotate2D file, couldn't annotate as it didn't show red dots, I was wondering if it is because I am running under windows, or maybe the python environment is not completely matching in terms of packages or their versions..., it is showing the picture and environment but when I click nothing happens.

leilaUEA commented 3 years ago

regarding using other models, is there any other 2D pose estimation frameworks have the same 2d skeleton format? for example openpose body_25 and COCO models doesn't have the extra thorax and spine I think, and your 2D model is more accurate having two extra keypoint. Is there any other package that produces this kind of skeleton? does this model of skeleton have any name I can look for?

Thank you very much, appreciate your help

Nicholasli1995 commented 3 years ago

Hi, Is there a possibility to do 2D pose estimation of an arbitrary image with the current code that is uploaded? the inference code is uploading skeleton from a fixed, predefined dictionary of annotation.... Thanks very much in advance Best regards, Leila

Hi, now the 2D pose estimation model is uploaded and the usage is available at https://github.com/Nicholasli1995/EvoSkeleton/blob/master/docs/2DHPE.md.

Nicholasli1995 commented 3 years ago

Thanks very much for your response and for uploading the 2D pose estimator, I appreciate it!

I also tried running the annotator, annotate2D file, couldn't annotate as it didn't show red dots, I was wondering if it is because I am running under windows, or maybe the python environment is not completely matching in terms of packages or their versions..., it is showing the picture and environment but when I click nothing happens.

Is there any error message?

Nicholasli1995 commented 3 years ago

regarding using other models, is there any other 2D pose estimation frameworks have the same 2d skeleton format? for example openpose body_25 and COCO models doesn't have the extra thorax and spine I think, and your 2D model is more accurate having two extra keypoint. Is there any other package that produces this kind of skeleton? does this model of skeleton have any name I can look for?

Thank you very much, appreciate your help

I don't know other pre-trained model that can do H36M style 2D pose estimation for in-the-wild images. You may try my uploaded model first. If it does not work, one can try training a different 2D pose estimation model or 2D-to-3D lifting model.

sidharthrai commented 3 years ago

@Nicholasli1995 Hi thank you for uploading your model. I was wondering where are images with the skeleton going? I can not find any after I run your pre-trained model.

Nicholasli1995 commented 3 years ago

@Nicholasli1995 Hi thank you for uploading your model. I was wondering where are images with the skeleton going? I can not find any after I run your pre-trained model.

Hi, I did not save them automatically since these images are just examples. You can save them yourself by running plt.savefig.

Nicholasli1995 commented 3 years ago

Closing inactive issues that already have answers. Re-open if needed.