{"full_name":"DeepLabCut/DeepLabCut","owner":"DeepLabCut","name":"DeepLabCut","shard":"1","url":"https://github.com/DeepLabCut/DeepLabCut","description":"Official implementation of DeepLabCut: Markerless pose estimation of user-defined features with deep learning for all animals incl. humans","homepage":"http://deeplabcut.org","topics":["behavior-analysis","deep-learning","pose-estimation","feature-detectors","toolbox","deeplabcut","animal-pose-estimation","labeling-tool","keypoint-tracking","keypoint-detection"],"tag_pages":["deep-learning","pose-estimation","toolbox"],"language":"Python","license":"LGPL-3.0","avatar":"https://avatars.githubusercontent.com/u/59848143?v=4","created_at":"2018-03-26T22:29:12Z","pushed_at":"2026-11-14T11:36:15Z","metrics":{"stars":5824,"forks":1797,"watchers":119,"issues":2197,"prs":1358,"releases":63,"issues_open":26,"issues_closed":2171,"prs_open":30,"prs_closed":182,"prs_merged":1146,"discussions":0},"growth":{"day":{"stars":2,"forks":0,"base_day":"2026-11-15"}},"history":{"dates":["2026-11-10","2026-11-11","2026-11-12","2026-11-13","2026-11-14","2026-11-15","2026-11-16"],"stars":[5817,5818,5820,5820,5822,5822,5824],"forks":[1798,1798,1797,1797,1797,1797,1797],"watchers":[null,119,119,119,119,119,119],"issues_open":[null,25,26,26,26,26,26],"prs_open":[null,26,29,29,32,30,30]},"boards":{},"same_owner":[],"issues":[{"number":3579,"title":"0% GPU utilization; `torch.cuda.is_available()` reports `true`","comments":6,"created_at":"2026-11-04T00:54:53Z"},{"number":2887,"title":"Unsolicited advice on packaging and compatibility","comments":5,"created_at":"2025-02-17T21:45:58Z"},{"number":3462,"title":"DLC Freezing Issue","comments":4,"created_at":"2026-08-25T19:43:38Z"},{"number":3281,"title":"[FR] Add an option to add addtitional individuals to multianimal model","comments":4,"created_at":"2026-04-09T15:02:55Z"},{"number":3585,"title":"validation errors for ProjectConfig while calibrating for 3D project","comments":3,"created_at":"2026-11-12T12:08:38Z"},{"number":3562,"title":"Cannot create labelled videos on Windows from videos analyzed on Linux","comments":3,"created_at":"2026-10-22T23:44:53Z"},{"number":3381,"title":"Update (or remove) command line interface","comments":3,"created_at":"2026-06-24T14:23:04Z"},{"number":3556,"title":"Extra inputs are not permitted","comments":2,"created_at":"2026-10-16T16:35:40Z"}],"discussions":[],"latest_release":{"name":"v3.0.2","tag":"v3.0.2","published_at":"2026-09-23T10:46:26Z","body":"# DeepLabCut 3.0.2\r\n\r\n\r\nDeepLabCut 3.0.2 is a patch release bringing several PyTorch workflows fixes,\r\nimproved config handling and export reliability, broadened compatibility for newer\r\nNumPy / Matplotlib versions, and adding a machine-readable docs/API index for\r\nagent-assisted interactions with the library.\r\n\r\n\r\n## Highlights\r\n\r\n- Forward-compatibility work for NumPy 2, pandas 3 CoW, and Matplotlib 3.9,\r\n  including releasing core NumPy / Matplotlib upper bounds (#3420, #3498,\r\n  #3502, #3508, #3499, #3445, #3394).\r\n- More reliable COCO training: `model_cfg.metadata` is authoritative, with\r\n  fail-fast validation against annotation JSON (#3435).\r\n- Safer legacy project configs: empty / `null` top-level YAML fields are treated\r\n  as unset; `ignore_empty` is removed (#3439).\r\n- Normalized `export_model` pickles for downstream / real-time consumers (#3473).\r\n- Detector `null` thresholds resolve to field defaults instead of crashing\r\n  (#3470, #3475).\r\n- PyTorch `analyze_videos` / auto-track honor the `overwrite` argument instead of\r\n  hard-coding `overwrite=False` (#3514).\r\n- Public LLM-friendly knowledge index (`llms.txt`, `docs.jsonl`, `api.jsonl`)\r\n  plus hardening follow-up and `AGENTS.md` (#3453, #3494, #3483).\r\n\r\n\r\n## Important changes\r\n\r\n### NumPy / Matplotlib dependency bounds (#3420, #3498, #3502, #3508, #3499)\r\n\r\n\r\nCore installs no longer pin `numpy<2` or a Matplotlib upper bound. Also added\r\n`numba>=0.60` and an explicit OpenCV dependency. TensorFlow extras still keep a\r\nNumPy `<2` ceiling where the corresponding TensorFlow build requires it.\r\n\r\n\r\n- #3420 — Prepare for NumPy 2 forward compatibility (including dense SVD for\r\n  tracklet ranking instead of randomized interpolative SVD)\r\n- #3498 — Fix broken NumPy 2.0 compatibility in `crossvalutils` and\r\n  `predict_multianimal`\r\n- #3502 — NumPy 2 + numba: take the real part of eig eigenvectors in\r\n  `EllipseFitter`\r\n- #3508 — Follow-up NumPy 2 fixes (PyTorch snapshot metric loading /\r\n  `Assembly.affinity` typing; raise `numba` floor; declare OpenCV)\r\n- #3499 — Remove core NumPy and Matplotlib upper bounds (NumPy `<2` remains\r\n  only where TensorFlow extras still require it; exclude broken Matplotlib\r\n  3.7 / 3.7.1)\r\n- #3445 — Guard pandas 3 Copy-on-Write read-only `.to_numpy()` / `.values`\r\n  mutations\r\n- #3394 — Matplotlib v3.9 compatibility\r\n- #3509 — Tighten strict canary CI handling (`strict_cells` for same-repo\r\n  pushes/PRs; fork PRs stay non-blocking)\r\n\r\n\r\n### COCOLoader (#3435)\r\n\r\n\r\n`bodyparts` / individuals capacity now come from `model_cfg.metadata`, matching\r\n`DLCLoader`. Construction validates `train.json` and `test.json` and raises\r\n`ValueError` if an image exceeds configured capacity or bodyparts do not match.\r\nProjects that previously relied on capacity inferred only from the busiest\r\ntraining image should set `max_individuals` (and bodyparts) explicitly.\r\n\r\n\r\n### ProjectConfig empty / null fields (#3439)\r\n\r\n\r\nEmpty top-level YAML fields are treated as unset and emit an actionable warning.\r\n`DLCBaseConfig.from_yaml(..., ignore_empty=...)` no longer accepts `ignore_empty`\r\n(breaking for that optional parameter). Lower-level `read_config(..., ignore_empty=...)`\r\ndeprecates the argument.\r\n\r\n\r\n### Exported models (#3473)\r\n\r\n\r\n`export_model` now calls `PoseConfig.to_dict(normalize=True)`, so pickles store\r\nprimitives (`Path` → `str`, enums → `str`, etc.). Prefer re-exporting models for\r\nconsumers that load the pickled config.\r\n\r\n\r\n### `overwrite` in PyTorch video analysis (#3514)\r\n\r\n\r\n`analyze_videos` and the auto-track path now forward the caller’s `overwrite`\r\nflag instead of always behaving as `overwrite=False`. Existing prediction files\r\nare skipped only when `overwrite` is false.\r\n\r\n\r\n## Features and enhancements\r\n\r\n\r\n- #3436 — Add `detector_eval_interval` to `train_from_coco`\r\n- #3475 — Add `DefaultIfNone` config validator (`null` → field default)\r\n- #3481 — Harden wandb logger error handling (validate before `wandb.init`)\r\n\r\n### Agentic interaction\r\n\r","is_prerelease":false},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-16T09:23:14Z","history_start":"2026-09-06"}