search

DirtyHarryLYL / Transferable-Interactiveness-Network

Code for Transferable Interactiveness Knowledge for Human-Object Interaction Detection. (CVPR'19, TPAMI'21)
MIT License
227 stars 41 forks source link

About human pose feats and HICO-Det Dataset Annotation #5

Closed YueLiao closed 5 years ago

YueLiao commented 5 years ago

Good job! 1.Do you use the alpha-pose result trained on MPII or COCO dataset? 2.I would like to know if you filter keypoints the score of which blow the threshold(maybe 0.05 in coco). If done, how to construct the human skeleton if missing some keypoints.

DirtyHarryLYL commented 5 years ago
  1. We used the pre-trained Alphapose model (on COCO).
  2. Yeah, we used the filtering. We have tried different filtering thresholds, e.g., 0.05, 0.1. The performances are similar. If some points are missing or discarded, the corresponding skeleton edges/lines will also be discarded. image
YueLiao commented 5 years ago

Thanks for the reply. I got it. Another question about hico-det detection annotation: On HICO-DET dataset, the same object may be annotated by several bouding-boxes which have high IOU with each other. I wonder if you have done something for it.

DirtyHarryLYL commented 5 years ago

Yes. If our setting, two human or object boxes with Iou>0.7 will be seen as the same person or object. We fuse the boxes which have high IoU，and merge their labels.

YueLiao commented 5 years ago

I got you. Thanks a lot. It's very helpful for me.