keypointfeature
wangjunbianqiang opened this issue · 4 comments
Hi,thanks your work! I want to know how to extract the keypoints features of the box pairs from the pictures.
Hi, there.
For more details on the pose features, please refer to Sec.3.2 in our paper.
I have a question, which is where did this N come from? The original key point feature was (number of person instance,17,2). How do I convert that to (number of box pairs,17,2)
N can be considered as batch size. During training, we calculated our box_pair (human-object) training data (Pose features) offline. We did not directly input the keypoint features into the model.
Can you tell me how that works? I've been doing a lot of research and I don't know how to relate keypoint features to box-pairs.
thank you very much!