Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

UDP for mmpose #228

Closed
HuangJunJie2017 opened this issue Nov 3, 2020 · 4 comments
Closed

UDP for mmpose #228

HuangJunJie2017 opened this issue Nov 3, 2020 · 4 comments
Assignees

Comments

@HuangJunJie2017
Copy link
Contributor

I am the author of UDP, and i am planing a new pull-request about UDP for mmpose
the main problem is that, to acquire the precision improvement, all the trained models in mmpose need to be retrained~
Some results list in our project for both bottom-up and top-down paradigm: https://github.com/HuangJunJie2017/UDP-Pose

@innerlee
Copy link
Contributor

innerlee commented Nov 3, 2020

Thanks! Looks fantastic! I have a few questions

...all the trained models in mmpose need to be retrained

Will hand/mesh tasks benefit from this? What is needed for the retrain, do we need to change config files, and any docs for instructions? The retrain happens after the pull request, and carried out by us, am I understanding correct?

@jin-s13
Copy link
Collaborator

jin-s13 commented Nov 3, 2020

Excited to hear that! PR is welcome.
I have read about UDP-Pose and I really like this work 👍 It is general and can boost the performance of both top-down and bottom-up methods.
Once the code is ready, we can help with training these models : )

@HuangJunJie2017
Copy link
Contributor Author

-Will hand/mesh tasks benefit from this?
UDP modifies the data processing pipeline which is used for all tasks. Theoretically speaking, hand/mesh tasks will benefit from this.
-What is needed for the retrain, do we need to change config files, and any docs for instructions?
For training, we just need to replace some code in 'mmpose/datasets/pipelines/bottom_up_transform.py' and 'mmpose/datasets/pipelines/top_down_transform.py' for bottom-up and top-down paradigm respectively.
For testing, we need to change some extra code about the post processing. Some configs may need to be modified, for example PROJECT2IMAG is no longer need for bottom-up paradigm as it is consumptive and doesn't provide any precision improvement.
I have not reviewed all this project and there may be something else need to be changed~
-The retrain happens after the pull request, and carried out by us, am I understanding correct?
Yes, this is consumptive

@HuangJunJie2017
Copy link
Contributor Author

@jin-s13 ok, i still need some time to review, prepare and test the code, i will pr in next week if it is ready

@jin-s13 jin-s13 closed this as completed Dec 15, 2020
HAOCHENYE added a commit to HAOCHENYE/mmpose that referenced this issue Jun 27, 2023
* fix runner dump cfg

* convert dict cfg to Config
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

No branches or pull requests

3 participants