KeyError: 'eye_distance', in regress_kp(batch_list), celeba_wild.py.

xingzhehe / AutoLink-Self-supervised-Learning-of-Human-Skeletons-and-Object-Outlines-by-Linking-Keypoints

[NeurIPS 2022] AutoLink, a simple and novel unsupervised approach to detect keypoints from single static images

https://xingzhehe.github.io/autolink/

MIT License

43 stars 7 forks source link

KeyError: 'eye_distance', in regress_kp(batch_list), celeba_wild.py. #7

Open LeonXu127 opened 1 month ago

LeonXu127 commented 1 month ago

In regress_kp(bctch_list) from celeba_wild.py, there are only 2 keys 'det_keypoints' and 'keypoints'. Should I add "'eye_distance': batch['eye_distance'].cpu()" to line 93 of model.py which is originally "out_batch = {'keypoints': batch['keypoints'].cpu(), 'det_keypoints': kp.cpu()}"? QQ_1720710315921

xingzhehe commented 1 month ago

Yes. You will need eye_distance during inference. I think it can also be inferred from the GT keypoints, but I seemed to forget to modify the dataset-specific testing code.