Replies: 1 comment 2 replies
-
Yes, during validation, the keypoints will be set to zero. The boxes are either from gt json (ann_file) or detection file (bbox_file). The trained model will predict the keypoints given the bbox. We will NOT merge keypoint annotations & detected bboxes. After that, during evaluation, we will reload the predictions and the gt annotations. |
Beta Was this translation helpful? Give feedback.
2 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Hi,
I'm training on a custom dataset (COCO foot) with a config file based on mspn50_coco_256x192.py. I've expected that the loader merges keypoint annotations (
ann_file
) and detected bboxes (bbox_file
).Unlike the expectation only the bboxes are loaded with all keypoints set to zero. The logic is quite clear in
mmpose/mmpose/datasets/datasets/top_down/topdown_coco_dataset.py
Line 98 in c1460e4
How should be the validation dataset correctly loaded? Thanks!
Beta Was this translation helpful? Give feedback.
All reactions