update pytorch models following DeepLabCut 3.0.0rc13#151
update pytorch models following DeepLabCut 3.0.0rc13#151MMathisLab merged 4 commits intoDeepLabCut:mainfrom
Conversation
This commit Updates the DEKR predictor to follow the DeepLabCut implementation in version 3.0.0rc7, see DeepLabCut/DeepLabCut#2907
This commit updates the DEKRPredictor to follow the DeepLabCut implementation in version 3.0.0rc13. see DeepLabCut/DeepLabCut#3121
This commit updates the PAF predictor to follow the DeepLabCut implementation in version 3.0.0.rc13. See DeepLabCut/DeepLabCut#3117
This commit updates the `HeatmapPredictor` in single_predictor.py to follow the implementation in DeepLabCut 3.0.0rc13. See DeepLabCut/DeepLabCut#3110
|
indeed the maintenance burden is real - would probably be worth it to set up a CI action that just diffs that directory in the source package with dlc-live and generates a PR, seems like would net immediate gains |
|
@sneakers-the-rat, agreed! We're still discussing about the best solution here. |
|
@C-Achard, we can add notes/tags inside the codebase where duplicated from DeepLabCut. Maybe in a separate PR. This one needs merging first. |
Is it unclear which modules are duplicated? I assumed the whole (Chiming in recently because we still use this package in tf mode but would be interested in swapping to torch if its as stable as the tf has been, so am interested in helping w maintainance) |
C-Achard
left a comment
There was a problem hiding this comment.
Looks good, I will test this a bit more but overall the changes are solid and should speed the pipeline up.
Motivation
TypeError: DEKRPredictor: DEKRPredictor.__init__() got an unexpected keyword argument 'nms_threshold')Changed models:
dlclive/pose_estimation_pytorch/models/predictors/dekr_predictor.py
(see NMS for DEKR models DeepLabCut#2907 and PyTorch: Speed up DEKR predictor DeepLabCut#3121)
dlclive/pose_estimation_pytorch/models/predictors/paf_predictor.py
(see PyTorch: Speed up PAF cost computation DeepLabCut#3117)
dlclive/pose_estimation_pytorch/models/predictors/single_predictor.py
(see PyTorch: Speed up single animal models DeepLabCut#3110)