search

jin-s13 / COCO-WholeBody

ECCV2020 paper "Whole-Body Human Pose Estimation in the Wild"
768 stars 73 forks source link

tensorflow implementation #21

Closed soans1994 closed 3 years ago

soans1994 commented 3 years ago

hello author,

thank you for the full body dataset. can this dataset be used to train a model with tensorflow? will it slow down the realtime performance in terms of fps?

thank you

jin-s13 commented 3 years ago

I have not used tensorflow before, but I think it would be easy to train the model using tensorflow.

soans1994 commented 3 years ago

thank you. which framework did you use for implementation.

jin-s13 commented 3 years ago

We use pytorch for implementation, the MMPose codebase. Please check it.

soans1994 commented 3 years ago

thank you. i will check it

soans1994 commented 3 years ago

while creating the whole body annotations for coco dataset, did you use landmark keypoint face detection model to extract the facial points? or did you manual annotation. and what about the foot annotations.

jin-s13 commented 3 years ago

For face/hand keypoints, we first use models pre-trained on a collection of different datasets to obtain psudo-labels, then we manually correct the mislabeled keypoints. For foot keypoints, we directly manually label the keypoints. More details can be found in the paper.

soans1994 commented 3 years ago

thank you very much for your response

soans1994 commented 3 years ago

@jin-s13
i have some doubts regarding face keypoints. can you please look into it. https://github.com/jin-s13/COCO-WholeBody/issues/22

thank you

soans1994 commented 2 years ago

hello again,

when i use the face keypoints from whole body json, i get some error for a sample in between, like SystemError: tile cannot extend outside image Is there some problem with the bbox values? Other works good all 68 landmarks. did any other people get similar error.

Thank you