Add 3.0 changelog#3340
Merged
Merged
Conversation
### Contents - Finalized 3.0 changelog - Adds benchmark images ### TODO - [ ] Fix image links to point to github user content on main
Replace relative image paths with raw.githubusercontent.com absolute URLs in changelog/3_0_0/v3_0_0.md so images render correctly when viewed in releases. Updated three image references (openfield_benchmark_pr2613.png, speed_tensorflow.avif, buctd_benchmarks.png).
deruyter92
approved these changes
May 21, 2026
C-Achard
added a commit
that referenced
this pull request
May 21, 2026
* version bump 3.0 * update uv.lock (deeplabcut version 3.0) * Add 3.0 changelog (#3340) * Add 3.0 changelog ### Contents - Finalized 3.0 changelog - Adds benchmark images ### TODO - [ ] Fix image links to point to github user content on main * Use raw GitHub URLs for changelog images Replace relative image paths with raw.githubusercontent.com absolute URLs in changelog/3_0_0/v3_0_0.md so images render correctly when viewed in releases. Updated three image references (openfield_benchmark_pr2613.png, speed_tensorflow.avif, buctd_benchmarks.png). --------- Co-authored-by: Cyril Achard <cyril.achard@epfl.ch> Co-authored-by: Mackenzie Mathis <mathis@rowland.harvard.edu>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Contents
TODO
Preview:
DeepLabCut 3.0: familiar workflows, modern foundations, better performance
DeepLabCut 3.0 introduces a PyTorch-first training and inference stack while keeping the core DeepLabCut workflow familiar.
Projects still follow the same labeling, training, evaluation, and video-analysis pipeline used throughout the 2.x series, but the underlying engine has been substantially modernized.
For users who have already been following the release candidates, many of these changes will already feel familiar.
DeepLabCut 3.0 consolidates these incremental changes into a stable release.
Increased model performance and speed
The journey to 3.0
A quick recap of some of the major milestones leading to this release:
And more, find the full PR reference on GitHub!
Notable features in 3.0.0
PyTorch-first, TensorFlow-compatible
DeepLabCut 3.0 adds a new PyTorch backend while retaining TensorFlow support for legacy workflows.
Project management remains the same and labeled datasets remain compatible.
PyTorch models can be trained alongside previous TensorFlow models on the same train/test splits for direct benchmarking and comparison.
Expanded architecture support
DeepLabCut 3.0 significantly broadens the supported model ecosystem beyond the classic ResNet-based workflows. The PyTorch stack includes:
The documentation now includes dedicated architecture guides to help users choose models based on scene complexity and experimental needs.
Flexible PyTorch training configuration
The PyTorch engine introduces a modern training stack with expanded augmentation options, training schedules, device management, and model architectures. For each training run, the settings are stored in a
pytorch_config.yaml, enabling easy reproducibility.Improved interoperability
The new PyTorch data pipeline introduces loaders for both standard DeepLabCut projects and COCO-style datasets, making it easier to integrate DeepLabCut with broader computer-vision workflows and external annotation formats.
Model Zoo and SuperAnimal workflows
DeepLabCut 3.0 continues to expand the Model Zoo and SuperAnimal ecosystem, making pretrained models and transfer learning more accessible.
Colab notebooks and updated GUI tooling make it easier to experiment with modern architectures without extensive setup. (see the documentation)
Modernized installation and packaging
The project has been moved to a newer packaging system, and is now based around pyproject.toml. This enables the use of modern package-managers & dependency resolvers, such as
uvorpdm.Users can still install only the components they require, be it GUI support, TensorFlow compatibility, ModelZoo features, and optional experimental integrations.
Labeling GUI
DeepLabCut 3.0 is shipped with a new release of the napari-deeplabcut plugin. Our napari-based labeling GUI has undergone a major internal re-write and modernization: while preserving familiar UI and the DeepLabCut workflow, the update substantially improves stability, data handling, usability, visualization, and annotation workflows, now with automated point tracking for faster labeling. See the release notes to find out about all improvements.
Upcoming: refreshed documentation
We have updated and streamlined the documentation, with a focus on clarity and up-to-date information in core areas (installation, getting started guides, and more).
Expect the documentation to continue evolving soon after the release!
A major transition
The jump from the final DeepLabCut 2.x releases to the current codebase is best understood as a transition to more recent Python & deep learning ecosystems rather than a routine update.
Taken together, the PyTorch backend, broader architecture support, ModelZoo integration, packaging modernization, updated labeling GUI, and documentation improvements represent a major evolution of DeepLabCut, which we are happy to release as 3.0.
Closing thoughts
We hope you enjoy this new version, and we aim to keep sharing many exciting improvements in the future in all areas, be it performance and speed, codebase quality improvements, foundation models integration, user experience and documentation.
Changelog since 3.0.0rc14
GenerativeSamplervisibility-aware #3305)@torch.inference_modefor inference on AMD DirectML GPUs (Add conditional replacement of@torch.inference_modefor inference on AMD DirectML GPUs #3295)list_videos_in_folder(updatelist_videos_in_folder#3303)TrainingDatasetMetadataandget_shuffle_enginefor incomplete projects (ImproveTrainingDatasetMetadataandget_shuffle_enginefor incomplete projects #3313)SimCCPredictor: exposeapply_softmaxand fix visibility thresholding (update RTMPoseSimCCPredictor: exposeapply_softmaxand fix visibility thresholding #3306)deeplabcut-dockerpackage #3291)