Pose-assisted behavior analysis
BehaviorScope-X#
Turn animal-behavior recordings into reviewable temporal predictions while keeping annotation, data splits, pose features, model training, and evaluation visible at every step.
One interface, three pose paths#
Choose the representation that matches your study. Every path returns to the same temporal behavior workflow, so annotation, training, held-out testing, and output review remain comparable.
YOLO-pose
Build caches, train the classifier, run inference, and create annotated review videos from the GUI-centered path.
Open workflow →MobileNetV3
Use a controlled pose-and-backbone descriptor path with shared temporal modeling and evaluation.
Open workflow →DeepLabCut-HRNet
Stage SuperAnimal pose, HRNet feature caches, classifier training, and held-out evaluation.
Open workflow →Research workflow#
The map follows the primary path from research recordings to evidence you can inspect. Open it separately for pan, zoom, and a larger canvas.
Workflow map generated with Archify, an MIT-licensed interactive diagram renderer. Special thanks to Shaun Andrade for introducing the project to Archify and for assisting with pilot studies and architecture testing during the early builds of BehaviorScope. See Acknowledgments.
Keep the research decisions visible#
Annotate bouts, record review state, and keep train, validation, held-out test, and excluded videos explicit.
Select a supported pose path, build compatible caches, and evaluate the temporal classifier on held-out data.
Inspect prediction tables, ethograms, bout summaries, metrics, and optional annotated videos before interpretation.
Special thanks#
BehaviorScope-X gratefully acknowledges Shaun Andrade, an undergraduate Computer Science major at the University of Maryland, Baltimore County (UMBC), for assisting with pilot studies and architecture testing during the early builds of the BehaviorScope project. Shaun also introduced the project to Archify, the workflow-diagram repository used to generate this site's interactive research map.
Start with the guided desktop workflow#
Install the application, annotate a first video, and follow one documented pose path through model training and output review.