DeepLabCut 3.0.2

DeepLabCut 3.0.2 is a patch release bringing several PyTorch workflows fixes,
improved config handling and export reliability, broadened compatibility for newer
NumPy / Matplotlib versions, and adding a machine-readable docs/API index for
agent-assisted interactions with the library.

Highlights

• Forward-compatibility work for NumPy 2, pandas 3 CoW, and Matplotlib 3.9,
  including releasing core NumPy / Matplotlib upper bounds (#3420, #3498,
  #3502, #3508, #3499, #3445, #3394).
• More reliable COCO training: model_cfg.metadata is authoritative, with
  fail-fast validation against annotation JSON (#3435).
• Safer legacy project configs: empty / null top-level YAML fields are treated
  as unset; ignore_empty is removed (#3439).
• Normalized export_model pickles for downstream / real-time consumers (#3473).
• Detector null thresholds resolve to field defaults instead of crashing
  (#3470, #3475).
• PyTorch analyze_videos / auto-track honor the overwrite argument instead of
  hard-coding overwrite=False (#3514).
• Public LLM-friendly knowledge index (llms.txt, docs.jsonl, api.jsonl)
  plus hardening follow-up and AGENTS.md (#3453, #3494, #3483).

Important changes

NumPy / Matplotlib dependency bounds (#3420, #3498, #3502, #3508, #3499)

Core installs no longer pin numpy<2 or a Matplotlib upper bound. Also added
numba>=0.60 and an explicit OpenCV dependency. TensorFlow extras still keep a
NumPy <2 ceiling where the corresponding TensorFlow build requires it.

• #3420 — Prepare for NumPy 2 forward compatibility (including dense SVD for
  tracklet ranking instead of randomized interpolative SVD)
• #3498 — Fix broken NumPy 2.0 compatibility in crossvalutils and
  predict_multianimal
• #3502 — NumPy 2 + numba: take the real part of eig eigenvectors in
  EllipseFitter
• #3508 — Follow-up NumPy 2 fixes (PyTorch snapshot metric loading /
  Assembly.affinity typing; raise numba floor; declare OpenCV)
• #3499 — Remove core NumPy and Matplotlib upper bounds (NumPy <2 remains
  only where TensorFlow extras still require it; exclude broken Matplotlib
  3.7 / 3.7.1)
• #3445 — Guard pandas 3 Copy-on-Write read-only .to_numpy() / .values
  mutations
• #3394 — Matplotlib v3.9 compatibility
• #3509 — Tighten strict canary CI handling (strict_cells for same-repo
  pushes/PRs; fork PRs stay non-blocking)

COCOLoader (#3435)

bodyparts / individuals capacity now come from model_cfg.metadata, matching
DLCLoader. Construction validates train.json and test.json and raises
ValueError if an image exceeds configured capacity or bodyparts do not match.
Projects that previously relied on capacity inferred only from the busiest
training image should set max_individuals (and bodyparts) explicitly.

ProjectConfig empty / null fields (#3439)

Empty top-level YAML fields are treated as unset and emit an actionable warning.
DLCBaseConfig.from_yaml(..., ignore_empty=...) no longer accepts ignore_empty
(breaking for that optional parameter). Lower-level read_config(..., ignore_empty=...)
deprecates the argument.

Exported models (#3473)

export_model now calls PoseConfig.to_dict(normalize=True), so pickles store
primitives (Path → str, enums → str, etc.). Prefer re-exporting models for
consumers that load the pickled config.

overwrite in PyTorch video analysis (#3514)

analyze_videos and the auto-track path now forward the caller’s overwrite
flag instead of always behaving as overwrite=False. Existing prediction files
are skipped only when overwrite is false.

Features and enhancements

• #3436 — Add detector_eval_interval to train_from_coco
• #3475 — Add DefaultIfNone config validator (null → field default)
• #3481 — Harden wandb logger error handling (validate before wandb.init)

Agentic interaction

Coding agents and LLM tools can now discover DeepLabCut knowledge without
scraping HTML.

CI builds and deploys a structured index with the docs site:

• llms.txt — llmstxt.org-style entry point at the docs
  root (published here
  after deploy)
• knowledge/manifest.json — which API versions are indexed and which is current
• knowledge/main/docs.jsonl — one JSON object per user-docs page and heading
• knowledge/<version>/api.jsonl — one JSON object per documented API symbol

Also added: AGENTS.md,
shared guidance for humans and coding agents.

• #3453 — Knowledge indexing generator + CI deploy (tools/knowledge_indexing)
• #3494 — Follow-up hardening for the LLM knowledge index (dependency bounds,
  stable vs main API links in llms.txt, safer skip/empty-dir publish
  behavior, anchor/regression fixes)
• #3483 — Add AGENTS.md for AI-assisted development

Bug fixes

• #3419 — Skip TensorFlow-backed tests cleanly when the tf extra is absent
• #3435 — Fix COCOLoader: treat model_cfg.metadata / max_individuals as
  authoritative instead of inferring solely from train.json
• #3439 — Update None-type validation for legacy ProjectConfig templates;
  remove ignore_empty
• #3441 — GUI config editor crash + shared deferred reload on DefaultTab
• #3447 — Evaluation visualization fixes, clean-up, and coverage improvements
• #3449 — Fix frame extraction failure with pathlib video selections
• #3452 — Fix incorrect SuperAnimal HumanBody rename mapping in ModelZoo /
  Hugging Face download
• #3467 — Fix faulty str / Path conversions
• #3468 — Fix latent bugs and misleading errors in GUI create_videos
• #3469 — Fix WandbLoggerConfig so extra wandb kwargs validate (scoop into
  wandb_kwargs)
• #3470 — Fix Faster R-CNN detector inference crashing on
  box_score_thresh: null
• #3473 — Fix export_model dumping unnormalized objects in pickle
• #3480 — Fix GUI tests: use config_path after rename
• #3484 — Fix tools imports under a plain pytest invocation
• #3486 — Docs: fix dead Model Zoo link in examples/README.md
• #3514 — Fix hard-coded overwrite=False in PyTorch analyze_videos /
  auto-track path

Please let us know if you face any issues with this release, especially related
to the latest forward compatibility efforts, we appreciate any issues or reports!

New Contributors

• @natsukium made their first contribution in #3417
• @AxelNoun made their first contribution in #3445
• @rysk-t made their first contribution in #3470
• @wilmund made their first contribution in #3486
• @michael-denyer made their first contribution in #3502

Full Changelog: v3.0.1...3.0.2

DeepLabCut 3.0.1 - a stability release with a major documentation overhaul and a new structured configuration system.

Highlights

• Comprehensive documentation audit — The user guide has been thoroughly reviewed, updated, and improved. Please check out:

  • Our updated user guide, which is restructured for better readability and updated to reflect the latest API and features.
  • Our brand-new developer documentation site, which provides full API references, architecture guides, and developer onboarding material.

• Structured & validated configurations — The config system has been migrated to typed Pydantic models. Existing config.yaml files remain fully compatible, but configs are now schema-validated on load with clear error messages for invalid entries. Some user-benefits include:

  • ProjectConfig: load and save your config.yaml with full schema validation — typos, wrong types, or invalid values are caught immediately with clear error messages.
  • Pretty-printing: call cfg.print() or pretty_print(cfg) to inspect any config in a readable tree.
  • Change-tracking: DLCVersionedConfig tracks which fields have been modified since the config was loaded from YAML, so when saving back to disk, log_changes() can report exactly what was changed.
  • Dict-like compatibility: existing code that accesses config keys like a dict still works.
  • Auto-aliasing of old field names: renamed keys are silently mapped, so older config files load without manual edits.

What's Changed

• Update deeplabcut-docker - version 0.1.0 by @deruyter92 in #3345
• Fix NaN handling in PredictKeypointIdentities postprocessor by @deruyter92 in #3378
• Migration to structured and validated configs by @deruyter92 in #3198
• Remove unused create_folder function from utils.py by @juan-cobos in #3359
• Refactor VideoProcessor class by @C-Achard in #3358
• Prepare migration to pandas 3.0 by @deruyter92 in #3360
• Fix SkeletonBuilder initialization from GUI by @C-Achard in #3380
• Add GUI extra in auto-updater; support uv by @C-Achard in #3346
• Docs audit [April 2026]: Full version by @C-Achard in #3341
• Migrate to pathlib.Path instead of str (first phase) by @deruyter92 in #3350
• Bump idna from 3.13 to 3.15 by @dependabot[bot] in #3337
• CI: Introduce functional scripts dependencies in CI test selection by @C-Achard in #3396
• Bump tornado from 6.5.5 to 6.5.7 by @dependabot[bot] in #3373
• Docs audit [April 2026]: Add dev docs by @deruyter92 in #3336
• Docs: Add links to dev docs in main user guide by @C-Achard in #3412
• Bump msgpack from 1.1.2 to 1.2.1 by @dependabot[bot] in #3377
• Bump soupsieve from 2.8.3 to 2.8.4 by @dependabot[bot] in #3388
• Docs: add CSPNext to pytorch architectures, minor formatting by @C-Achard in #3413
• Miscellaneous imports and code fixes by @C-Achard in #3411
• GUI: Fix redundant signal calls, streamline path selections by @C-Achard in #3395
• Fix PySide6 stylesheet usage by @C-Achard in #3371
• Refactor ctd_conditions / conditions_provider + add ConditionsConfig + fix InferenceConfig by @deruyter92 in #3387
• bump torch to version 2.13 (update uv.lock only) by @deruyter92 in #3400
• GUI: Config validation errors handler dialog & layout fixes by @C-Achard in #3397
• Update deprecation markers (3.1 -> 3.0.1) by @deruyter92 in #3422
• Harden path-handling after pathlib migration by @deruyter92 in #3421
• version bump 3.0.1 by @deruyter92 in #3423

Full Changelog: v3.0.0...v3.0.1

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

Pose estimation performance of the 3.0 PyTorch models compared against previous TensorFlow models on the DeepLabCut Openfield dataset (see PR #2613); RMSE: root mean squared error. *Values from Mathis et al. 2018.

Speed of the current PyTorch implementation (ResNet50) compared to the TensorFlow implementation. Results were obtained using a NVIDIA GeForce RTX 2080 Ti with CUDA 12.2 on the DeepLabCut Trimice dataset.

Comparison of the new BUCTD model architectures with DLCRNet and DEKR on the Marmoset, Fish and Trimice dataset. From Zhou et al., ICCV 2023.

The journey to 3.0

A quick recap of some of the major milestones leading to this release:

• #2613 - Full initial PyTorch backend implementation
• #2952 - New bottom-up conditional top-down (BUCTD) model architecture
• #2795 - New RTMPose top-down architecture
• #2868 - Updated notebooks and Colab examples for PyTorch workflows
• #2804 - PyTorch model export

And more, find the full PR reference on GitHub!

Credits to @n-poulsen, @jeylau, @maximpavliv, @MMathisLab, @yeshaokai, @LucZot, @AlexEMG, @arashsm79 and many other contributors.

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:

• ResNet and HRNet backbones
• Bottom-up multi-animal approaches such as DEKR and PAF/DLCRNet variants
• Top-down detector-plus-pose pipelines including RTMPose
• Hybrid architectures such as BUCTD and CTD variants
• SuperAnimal-related pretrained workflows

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 uv or pdm.
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

• DLC-live-GUI docs : Install from PyPI pre-release (#3238) by @C-Achard
• Apply linting to entire codebase & add CI workflow to check linting (#3216) by @deruyter92
• Bump requests from 2.32.5 to 2.33.0 (#3259) by @app/dependabot
• Refactor Analyze Videos tab (#3268) by @C-Achard
• Consolidate test workflow infrastructure in CI (#3254) by @C-Achard
• Move protobuf requirement to pyproject.toml (#3235) by @C-Achard
• Use pinned ffmpeg version in CI (#3276) by @C-Achard
• Docs versioning: Add glob support, better validation and reporting (#3278) by @C-Achard
• Bump cryptography from 46.0.5 to 46.0.7 (#3277) by @app/dependabot
• Fix failing local Windows tests due to ruamel parsing (#3275) by @C-Achard
• Update & de-duplicate skeleton builder (#3258) by @C-Achard
• Bump pygments from 2.19.2 to 2.20.0 (#3262) by @app/dependabot
• update conda yaml: add pyside6 as conda dependency (#3253) by @deruyter92
• Bump pyasn1 from 0.6.2 to 0.6.3 (#3249) by @app/dependabot
• Fix SuperAnimal / pretrained load for RTMPose: implement convert_weights on RTMCCHead (#3270) by @deruyter92
• Run update check in a separate thread (#3234) by @mikhailmikhalchuk
• Update napari-DLC docs for refactor (#3280) by @C-Achard
• Refactor/predict multianimal (#3220) by @juan-cobos
• Fix incorrect MultiLevel construction in outlier_frames.compute_deviations (#3247) by @C-Achard
• Bump pillow from 12.1.1 to 12.2.0 (#3283) by @app/dependabot
• Bump pytest from 9.0.2 to 9.0.3 (#3284) by @app/dependabot
• CircleCI: disable hugginface xet (#3316) by @deruyter92
• Update and diversify TensorFlow optional installations. (#3292) by @deruyter92
• Bump urllib3 from 2.6.3 to 2.7.0 (#3325) by @app/dependabot
• Bump gitpython from 3.1.47 to 3.1.50 (#3322) by @app/dependabot
• make CTD GenerativeSampler visibility-aware (#3305) by @deruyter92
• Add isatty method to StreamWriter + eval GUI fix (#3314) by @yyexela
• Add conditional replacement of @torch.inference_mode for inference on AMD DirectML GPUs (#3295) by @deruyter92
• Robustness fix: Annotation file not dropping likelihood column if present from machine labels (#3323) by @C-Achard
• Remove trailing comma in models_to_framework.json (#3330) by @C-Achard
• update list_videos_in_folder (#3303) by @deruyter92
• Improve TrainingDatasetMetadata and get_shuffle_engine for incomplete projects (#3313) by @deruyter92
• update RTMPose SimCCPredictor: expose apply_softmax and fix visibility thresholding (#3306) by @deruyter92
• GUI: Add "Generate debug log" action (#3328) by @C-Achard
• [Docker 1] Simplify and modernize Dockerfile (#3290) by @deruyter92
• [Docker 2] Update deeplabcut-docker package (#3291) by @deruyter92
• Add additional drop_likelihood_columns guards (#3333) by @C-Achard
• Resolve inconsistent parameter names via aliasing + depreca...

Highlights

• New installation procedure via pyproject.toml : easier, more modern installation. Enables uv, pdm and other modern package managers use.

• Many fixes and improvements:

  • PyTorch speed improvements
  • CI improvements

_{Note: these changes are based on all merges since rc10}

What's Changed

• link update for COLAB_YOURDATA_SuperAnimal.ipynb by @MMathisLab in #3037
• RTMpose_x superanimal_humanbody by @MMathisLab in #3043
• skeleton fix fo RTMpose demo by @MMathisLab in #3048
• Update COLAB_HumanPose_with_RTMPose.ipynb by @AlexEMG in #3051
• revert to working RTMPpose demo notebook by @MMathisLab in #3052
• fix for humanbody vs. others by @MMathisLab in #3050
• Save all pytorch config updates in dlc_torch.train_network() by @maximpavliv in #3062
• Save WandB run information to model folder by @maximpavliv in #3017
• Improve DLCLoader.to_coco() and add force_multi_animal argument to dlc_torch.evaluate() by @maximpavliv in #2995
• Superanimal humanbody: Use filtered detector by @maximpavliv in #3066
• ModelZoo GUI Tab fixes by @maximpavliv in #3055
• superanimal_humanbody with FilteredDetector: add video adaptation by @maximpavliv in #3086
• Expose close_figure_after_save in superanimal_analyze_images() by @maximpavliv in #3082
• Fix CTD empty conditions edge case by @maximpavliv in #3076
• Clean dlc_torch.analyze_videos() by @maximpavliv in #3079
• Fix RTMPose likelihood computation by @maximpavliv in #3078
• Fix typos in documentation and comments by @arashsm79 in #3097
• Fix snapshot index handling in PyTorch analyze_videos and extend overrides to other methods by @maximpavliv in #3092
• Updating version by @AlexEMG in #3096
• Improve threading termination and queue handling by @maximpavliv in #3094
• Fix deeplabcut.analyze_images() with CTD model by @maximpavliv in #2990
• Log GPU memory usage by @maximpavliv in #3077
• resize the GUI by @MMathisLab in #3034
• PyTorch: Improve inference batching speed by @arashsm79 in #3099
• Hide GPU memory usage by default by @maximpavliv in #3102
• Add autocast sub-config to inference settings and disable torch.autocast by default by @maximpavliv in #3105
• Move conditions: to inference: block and document inference-related configs by @maximpavliv in #3106
• Update TensorFlow backend for Python 3.12 compatibility (Keras 2 via tf-keras) by @maximpavliv in #3087
• new version by @AlexEMG in #3109
• Remove trailing underscore in superanimal scorer names by @maximpavliv in #3113
• update Argus documentation notes by @backyardbiomech in #3124
• Fix CI/CD failing on Mac by @arashsm79 in #3128
• PyTorch: Speed up single animal models by @arashsm79 in #3110
• PyTorch: Speed up PAF cost computation by @arashsm79 in #3117
• PyTorch: Speed up DEKR predictor by @arashsm79 in #3121
• Change bug report assignee by @MMathisLab in #3136
• docs: pin jupyter-book version by @arashsm79 in #3139
• Skip multiprocessing pytest by @deruyter92 in #3142
• Add argument max workers for multiprocessing in create_labeled_video by @deruyter92 in #3143
• Remove config editor delegate hack. by @deruyter92 in #3167
• Avoid redundant superanimal colormap calls during pytorch inference by @juan-cobos in #3159
• fix a confusing message for freeze_bn_stats by @Jiefei-Wang in #3166
• Line 355 output message updated address misleading error message for empty detections in analyze_images() function by @Ronniekev in #3157
• append, instead of rewriting, CSVLogger by @juan-cobos in #3179
• feat: Filter out low-confidence bounding box detections by @deruyter92 in #3154
• add missing init.py to pose_estimation_tensorflow/modelzoo. by @deruyter92 in #3153
• Fix GitHub Actions disk space exhaustion on Ubuntu runners by @Copilot in #3180
• Reverting PR #3179 which breaks csv logging. by @deruyter92 in #3183
• Update napari plugin activation in launch_napari by @C-Achard in #3181
• Refactor list_videos_in_folder for clarity and correctness by @juan-cobos in #3175
• Restrict pandas dependency to version <3.0 by @deruyter92 in #3186
• hotfix extract_outlier_frames for cropped videos by @deruyter92 in #3169
• fix non-deterministic unintended failing of test_coarse_dropout by @deruyter92 in #3182
• fix ruamel.yaml dumping newline for long filenames by @deruyter92 in #3140
• fix broken links in documentation by @deruyter92 in #3145
• Close figures before plotting by @deruyter92 in #3168
• Fix huggingface_hub and filelock dependency issue by @deruyter92 in #3204
• Fix CSVLogger: read existing learning_stats.csv file when resuming previous training by @deruyter92 in #3177
• Fix hardcoded GUI icon path by @deruyter92 in #3200
• Keyword-only key argument for pd.DataFrame.to_hdf() by @deruyter92 in #3185
• Replace str, os.path with pathlib.Path by @deruyter92 in #3199
• fix rename_mapping in download_huggingface_model (change arg type str -> dict) by @deruyter92 in #3196
• improve error handling in create_training_dataset by @deruyter92 in #3146
• Update docs TOC and Sphinx excludes by @C-Achard in #3201
• Add fmpose3d functionality by @deruyter92 in #3208
• Fix: properly use unique_bodyparts and identity_scores arguments in ShelfWriter.add_prediction by @maximpavliv in #3112
• Handle tables via pip as pandas extra instead of separate conda install (and update installation instructions) by @deruyter92 in #3214
• Update conda env yaml: bump DeepLabCut and remove nb_conda by @deruyter92 in #3213
• Install through pyproject.toml (PEP 518/621) by @deruyter92 in #3184
• Fix redundant/incomplete installs in docs CI, replace with pyproject.toml optional deps by @C-Achard in #3226
• Update CircleCI for pyproject.toml-based install by @C-Achard in #3225
• DeepLabCut-live-GUI PySide6/multicamera documentation ("2.0") by @C-Achard in #3206
• rc14 release by @AlexEMG in #3237

New Contributors

• @arashsm79 made their first contribution in #3097
• @juan-cobos made their first contribution in #3159
• @Jiefei-Wang made their first contribution in #3166
• @Ronniekev made their first contribution in #3157

Full Changelog: v3.0.0rc10...v3.0.0rc14

What's Changed

• Fix PySide6 AttributeError and uninitialized variable by @jeylau in #2622
• Niels/pcutoff arg by @n-poulsen in #2630
• Fix video overwrite by @jeylau in #2633
• Visually signal end of super animal video inference by @jeylau in #2637
• Fix GUI bugs by @jeylau in #2649
• Fix OSError: Could not find a backend to open CollectedData_Alex.csv with iomode r by @jeylau in #2650
• Fix video extension filter in filterpredictions by @jeylau in #2673
• Fix and decouple video rotation from other operations by @jeylau in #2653
• merge named snapshot evaluation into pytorch_dlc by @n-poulsen in #2657
• Handle internet connectivity issues by @jeylau in #2672
• New PyTorch DeepLabCut SuperAnimal demo notebook by @yeshaokai in #2629
• merge main to pytorch_dlc by @n-poulsen in #2677
• Update train.py: avoids error on 'unique_bodyparts' by @Zelin2001 in #2668
• Default PAF config update by @n-poulsen in #2678
• New detector options and better detector integration by @n-poulsen in #2676
• backwards compatibility by @n-poulsen in #2681
• evaluation refactor by @n-poulsen in #2679
• bug fixes for analyze_videos and unique_bodyparts by @Zelin2001 in #2685
• Fix import in deeplabcut.pose_tracking_pytorch by @jeylau in #2691
• Location Refinement for Top-Down models by @n-poulsen in #2699
• Identity Predictor by @n-poulsen in #2707
• batched inference by @n-poulsen in #2708
• Bug fixes: find_closest_neighbors + ellipse tracking by @n-poulsen in #2709
• bug fix: find closest neighbors by @n-poulsen in #2717
• Minimal backward compatibility by @AlexEMG in #2713
• Documentation Improvements - DeepLabCut PyTorch Code by @AlexEMG in #2714
• merge main into pytorch_dlc by @n-poulsen in #2731
• Make 3D Triangulate possible with pytorch_dlc branch and MA project by @Zelin2001 in #2729
• Data compatibility - Benchmark by @n-poulsen in #2740
• Adding cropping to pose_estimation_pytorch/apis/analyze_videos.py by @YankoFelipe in #2737
• pytorch_dlc - Small updates and bug fixes by @n-poulsen in #2747
• calc_similarity_with .py ZeroDivisionError: float division by zeros via upload by @SowonKIMM in #2688
• update version by @AlexEMG in #2748
• SuperAnimal Model Updates by @n-poulsen in #2756
• Look for uppercase video extensions when creating projects by @n-poulsen in #2763
• Fix reading gpus config by @maximpavliv in #2764
• Git ignore IDE configs by @maximpavliv in #2765
• Fix bug introduced in PR #2764 by @maximpavliv in #2772
• Bug fix: create_video_with_all_detections fails with PAF heads by @n-poulsen in #2776
• Resume training from existing snapshots option in GUI by @maximpavliv in #2774
• Improvements to the training UX by @n-poulsen in #2775
• Fix failing DLC 3.0 tests - changes to internal implementation by @n-poulsen in #2778
• Merge changes made to main by @n-poulsen in #2779
• Fix: gui/tabs/train_network.py, AttributeError: ‘NoneType’ object has no attribute ‘lower’ by @n-poulsen in #2780
• Use architecture-specific snapshots prefixes in video adaptation by @maximpavliv in #2781
• Video Analysis with Identity - Lowering memory footprint by @n-poulsen in #2785
• Timeout retrieving latest package version from the Web by @maximpavliv in #2782
• Load scheduler state when resuming training by @n-poulsen in #2788
• DeepLabCut 3.0 - Video Analysis with use_shelve=True by @n-poulsen in #2790
• Fix: correct the early return error when save_epochs=1 and optimize the clean up way by @xiu-cs in #2793
• Fix: Video Analysis for Single Animal projects by @n-poulsen in #2798
• Model Output Visualization for PyTorch Models by @n-poulsen in #2799
• Add Functions for Visualizing COCO Predictions and Ground Truth by @xiu-cs in #2792
• DLC 3.0 - Function to export PyTorch models by @n-poulsen in #2800
• KeypointAwareCrop: uniform sampling when no keypoints are present by @n-poulsen in #2801
• Fix version check timeout by @maximpavliv in #2805
• Revert "Fix version check timeout" by @n-poulsen in #2807
• PyTorch model evaluation - per_keypoint_evaluation and comparisonbodyparts by @n-poulsen in #2810
• Prune PAF graphs before evaluation by @n-poulsen in #2813
• Plot detector's bboxes if Top-Down architecture by @maximpavliv in #2791
• superanimal_analyze_images: improve plotting functions by @n-poulsen in #2815
• Update superanimal_analyze_images plotting by @n-poulsen in #2817
• Set weights_only=True as a default when loading snapshot weights. by @n-poulsen in #2818
• Implement missing arguments for PyTorch models by @n-poulsen in #2816
• Add missing positional arguments to SuperAnimal Colab by @maximpavliv in #2819
• New Top-Down Pose Estimation Model - RTMPose by @n-poulsen in #2795
• SuperAnimal demo colab update (for DLC3) by @MMathisLab in #2814
• Improved loading of snapshot weights with torch.load(..., weights_only=True) by @MMathisLab in #2823
• create_labeled_video fails with single animal TensorFlow models by @n-poulsen in #2826
• Merges main into pytorch_dlc by @n-poulsen in #2828
• Restore "Build skeleton" button by @maximpavliv in #2831
• Improve deeplabcut.pose_estimation_pytorch imports by @n-poulsen in #2833
• Improve train_network() docstring by @maximpavliv in #2836
• get_detector_inference_runner: resolve the device from the model config when None by @nattse in #2841
• Update docs video_inference_superanimal() by @maximpavliv in #2848
• Fix - setting cropping parameters in analyze_videos for PyTorch models by @n-poulsen in #2849
• Improvements to default configs by @n-poulsen in #2860
• Dynamic update of pytorch config with dot notation by @maximpavliv in #2859
• fix broken train_network argument display_iters in the GUI by @n-poulsen in #2865
• ReID Tracking dataset creation for PyTorch models by @n-poulsen in #2861
• Merges main into pytorch_dlc by @maximpavliv in #2876
• fix evaluation for ma models, following change in 'deeplabcut/core/metrics/api.py' by @Zelin2001 in #2872
• Fix check version timeout Version 2 (compatible with Windows and MacOS) by @maximpavliv in #2820
• fix entrypoint for dlc by @hmaarrfk in #2863
• Fix entrypoint - Requested changes by @maxi...

What's Changed

• Update beginners-guide.md by @MMathisLab in #2619
• Update DEEPLABCUT.yaml by @EitanHemed in #2625
• Add feature to evaluate network for specfic, named snapshots. by @Tetra-quark in #2508
• fixed file dialog not showing uppercase file extension video files by @park-jsdev in #2645
• Swap button added to the track refinement GUI by @JulianAlvarezdeGiorgi in #2680
• Fixes for testscript_3d.py by @jesse-gilmer in #2443
• Allow processing of videos in MKV containers by @ximion in #2701
• add in CZI EOSS badge and funding section by @MMathisLab in #2735
• Fix minimalnumberofconnections by @segashin in #2718
• Benchmarking user guide by @dikraMasrour in #2716
• adding docs for using data labeled elsewhere (formatting) by @MMathisLab in #2736
• Update standardDeepLabCut_UserGuide.md by @bendichter in #2761
• Create citation.md by @MMathisLab in #2767
• Fix failing tests - tmpdir_factory dir creation on OSX by @n-poulsen in #2768
• Update DLC3 PyTorch docs by @n-poulsen in #2804
• Pins numpy<2, fix svd for scipy>=1.11.0 by @n-poulsen in #2827
• Fix reading filtered state by @maximpavliv in #2844
• Remove pickle5 dependency from madlc Colab by @maximpavliv in #2846
• Fixed dropimagesduetolackofannotation to work with single-animal data by @n-poulsen in #2850
• Fixed conversion of dict to list for proper future indexing by @m-beau in #2855
• fixes to label_frames, addition of docstring by @n-poulsen in #2858
• Pin TensorFlow according to OS; Fix DeepLabCut install in Colab to deal with new requirements. by @n-poulsen in #2862
• Remove unused GUI button in analyze videos tab by @maximpavliv in #2875
• Fix video addition in VideoSelectionWidget by @maximpavliv in #2878
• v2.3.11 by @AlexEMG in #2882
• improved handling of animal names when analyzing videos by @n-poulsen in #2884

New Contributors

• @park-jsdev made their first contribution in #2645
• @JulianAlvarezdeGiorgi made their first contribution in #2680
• @jesse-gilmer made their first contribution in #2443
• @ximion made their first contribution in #2701
• @segashin made their first contribution in #2718
• @dikraMasrour made their first contribution in #2716
• @bendichter made their first contribution in #2761
• @m-beau made their first contribution in #2855

@maximpavliv joined the software team!

Funding includes ChanZuckerberg Initiative's Essential Open Source Software for Science. Thanks!

Full Changelog: v2.3.10...v2.3.11

What's Changed

• Doc re-arrangement in the TOC by @MMathisLab in #2523
• slight change to setup (move all TF to TF tag; pin tables, add note that python 3.9 is currently required by @MMathisLab in #2517
• Use video crop coordinates appropriately in create_video_with_all_detections by @jeylau in #2524
• Avoid infinite loop in mine() when there are no overlapping tracklets by @jeylau in #2525
• fix manual frame extraction with symlinks by @n-poulsen in #2527
• Update COLAB_DLC_ModelZoo.ipynb by @MMathisLab in #2533
• Adding a Quick Guide (from the wiki to the docs!) by @MMathisLab in #2532
• quick file fix by @MMathisLab in #2534
• Update installTips.md by @MMathisLab in #2536
• Video analysis on cropped videos: bug fixes for extract_outlier_frames and create_video_with_all_detections by @n-poulsen in #2538
• MacOS CI Fix - FFMpeg Installation by @n-poulsen in #2559
• niels/check_for_duplicate_videos by @n-poulsen in #2558
• Update setup.py by @MMathisLab in #2526
• bug fix: missing images during evaluation by @n-poulsen in #2570
• Update label and warn users of the absence of videos by @jeylau in #2586
• pin matplotlib version by @n-poulsen in #2583
• Correct DEEPLABCUT.yaml and DEEPLABCUT_M1.yaml reference. by @midumitrescu in #2573
• minor spelling fixes by @MMathisLab in #2594
• Fix 2d line data setter by @jeylau in #2599
• Auto check for dlclibrary & napari-deeplabcut updates when the GUI opens by @jeylau in #2600
• Update README.md by @MMathisLab in #2601
• Release by @AlexEMG in #2602

New Contributors

• @midumitrescu made their first contribution in #2573

Full Changelog: v2.3.9...v2.9.10

• Add GUI help button back by @jeylau in #2507

What's Changed

• Pull demo data from Zenodo by @jeylau in #2440
• Update maDLC_UserGuide.md by @Timokleia in #2446
• Fix bug dataset and metadataset fixes #2390 by @Tetra-quark in #2405
• Bug fix: add_new_videos(..., extract_frames=True) by @n-poulsen in #2468
• bug fix: empty predictions during video analysis by @n-poulsen in #2467
• Update README.md by @MMathisLab in #2464
• Update AUTHORS by @MMathisLab in #2475
• updated macOS runner to 13 by @n-poulsen in #2484
• Update README.md by @Timokleia in #2496
• Bump version of tables installed during CI by @jeylau in #2500
• updated COLAB notebooks to install a CUDA version compatible with DLC by @n-poulsen in #2506
• updated package requirements by @n-poulsen in #2513
• Test for the presence of snapshot files when using modelzoo by @jeylau in #2485
• Version update by @AlexEMG in #2514

Full Changelog: v2.3.8...v2.3.9

What's Changed

• better augmentation for maDLC: Fix fliplr augmentation for multi animal 2 by @biol-jsh in #2043
• Improved transfer learning for DLC ModelZoo by @yeshaokai in #2389
• Improvements to the model zoo's GUI tab by @jeylau in #2431
• "deeplabcut.filterpredictions" and "deeplabcut.analyze_skeleton" return mappings by @rahulporuri in #1908
• Let user change the search window size during corner detection refinement by @jeylau in #2414
• Install DLC from PyPI rather than its local version in M1 Macs by @jeylau in #2416
• Correct link to platform specifications page by @molecules in #2423
• Automatically check for napari-deeplabcut updates (in addition to DLC) by @jeylau in #2422
• Smartly restore pretrained model weights by @jeylau in #2426
• Fix video adaptation by @jeylau in #2436

New Contributors

• @molecules made their first contribution in #2423

Funding includes ChanZuckerberg Initiative's Essential Open Source Software for Science. Thanks!

Full Changelog: v2.3.7...v2.3.8

What's Changed

• update the lr when switching to adam optimizer by @n-poulsen in #2393
• Update maDLC_UserGuide.md by @Timokleia in #2394
• Fix testscript.py execution: set max_iters when training network by @n-poulsen in #2395
• Update python-package.yml by @MMathisLab in #2401
• Bump python to 3.9 in DEEPLABCUT.yaml environment file by @jeylau in #2402
• Address underlying compatibility issue with scipy 1.11.0 by @hmaarrfk in #2396
• Update standardDeepLabCut_UserGuide.md by @MMathisLab in #2403
• Unifying docs on Napari DeepLabCut by @Timokleia in #2399
• Update publish-book.yml by @MMathisLab in #2404
• Napari Flowchart by @Timokleia in #2406
• Minor modifications to deploy_book github action by @jeylau in #2407
• Update URL of napari shortcuts SVG by @jeylau in #2408
• update notebooks and COLAB by @n-poulsen in #2411
• GUI - Fix config YAML edition (start list indexing at 0) by @n-poulsen in #2412
• Fix link typos and conversion by @AlexEMG in #2409
• Update version by @AlexEMG in #2413

Full Changelog: v2.3.6...v2.3.7

Funding includes ChanZuckerberg Initiative's Essential Open Source Software for Science. Thanks!