search

luminxu / Pose-for-Everything

The official repo for ECCV'22 paper: Pose for Everything: Towards Category-Agnostic Pose Estimation
Apache License 2.0
197 stars 17 forks source link

confusing in train processing #11

Open Lavender0423 opened 2 years ago

Lavender0423 commented 2 years ago

Hi, thanks for your great work. I'm not so familiar with mmpose framework. Recently I trying to retrain the model by my own dataset of bird,I structure the annotation with coco standard and wtrite the config file according to the config file in your project,due to the dataset of bird has 15 keypoints,I had changed the channel_cfg like this image but,in trainning process I have got a high acc_pose and a low PCK in evaluation,like this image I'm confusing so I generate partial image in evaluate processing ,I notice that some keypoints are loacte at same position,can you tell me any reason about this？Is the config file wrong?or something in model need to change? preview image with keypoint and box is this image image

kltrock commented 2 years ago

I think there is something wrong with the implementation of their method. I observed the same behavior with their dataset too.

Lavender0423 commented 2 years ago

I think there is something wrong with the implementation of their method. I observed the same behavior with their dataset too.

/cry/cry

jin-s13 commented 4 months ago

Thank you for reporting this issue. I appreciate you taking the time to bring this to our attention. Unfortunately, the original author of this repository has since graduated and taken a position in industry. They no longer have plans to maintain the codebase going forward. If anyone in the community would be willing to help resolve this issue, I would be incredibly grateful. I would be happy to merge any fix that is proposed and tested.