RESEARCH USE
Where DeepLabCut fits
Define keypoints that answer the research question, label representative frames spanning poses, individuals, backgrounds, and acquisition conditions, then train and evaluate a pose model before processing videos in batches. Preserve source video, annotation revisions, training configuration, data splits, checkpoints, filtering and calibration settings, and manually review low-confidence, occluded, and out-of-domain segments.
Research tasks
- Estimate animal or object keypoints from video
- Analyze markerless pose for one or multiple animals
- Generate coordinate time series for kinematic and behavioral studies
What to evaluate before use
- Keypoint accuracy depends on annotation quality and training coverage. Occlusion, motion blur, camera changes, and new individuals can produce domain shift, so average error does not replace scene-level review.
- Pose coordinates are not behavior labels or biological interpretation. The core software is primarily LGPL-3.0, while SuperAnimal pretrained models have separate research-only, non-commercial terms.
Verification note
This entry summarizes the resource's role without assessing scientific accuracy or endorsing its outputs. Features and terms can change; consult the official source before adopting it for consequential work.
pose estimationbehavior analysis
Last verified: 2026-09-18
Source: official documentation ↗