I've ported TokenPose and TransPose into MMPOSE. #3174
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
I'm excited to share that I've had the opportunity to contribute a portion of code to the MMPOSE community. It's a privilege to collaborate with such a talented group of developers and contribute to this innovative project.
Motivation
Due to the overlap with my work, I have ported some keypoint detection code into the MMPOSE framework to meet my job requirements. I hope to contribute this portion of code directly to the community and learn about team-based workflow processes.
Modification
I have ported TokenPose and TransPose into the MMPOSE community. Here are the links to the papers and source codes.
TokenPose Paper Link:https://openaccess.thecvf.com/content/ICCV2021/html/Li_TokenPose_Learning_Keypoint_Tokens_for_Human_Pose_Estimation_ICCV_2021_paper.html
TokenPose Source Code:https://github.com/leeyegy/TokenPose
TransPose Paper Link:https://ieeexplore.ieee.org/document/9711155
TransPose Source Code: https://github.com/yangsenius/TransPose
My porting of the TokenPose and TransPose code to the MMPOSE community is based on the respective papers, with specific implementations primarily derived from the existing source codes. Both TokenPose and TransPose utilize HRNet for feature extraction, and I implemented this using the existing code in MMPOSE, without rewriting the source code from scratch. Furthermore, I have integrated the related model implementations into a single codebase, which has been fully implemented on my own COCO-type dataset.
However, I am not a computer science major, and my lab lacks sufficient computational resources. As a result, I have not conducted experiments on the COCO dataset but have completed my work on my own dataset. Additionally, I don’t fully understand aspects like bug detection and unit testing. That said, the MMPOSE community has provided me with significant help in accomplishing my work, and I sincerely want to give back by contributing to the community. I’m also very grateful to everyone who has dedicated their efforts to making the community better.
Use cases (Optional)
As the end of the year approaches, I've been overwhelmed with work and haven't managed to update the .md and .yaml files for TokenPose and TransPose. I apologize for this oversight.
Checklist
Before PR:
After PR: