Lightning Pose dataset: mirror-mouse

Head-fixed mice run on a circular treadmill while avoiding a moving obstacle (Warren et al., Elife 2021). The treadmill has a transparent floor and a mirror mounted inside at 45◦, allowing a single camera to capture two roughly orthogonal views (side view and bottom view via the mirror) at 250 Hz. 17 keypoints are labeled across the two views including seven keypoints on the mouse’s body per view, plus three keypoints on the moving obstacle.
The labeled data are partitioned into disjoint "in-distribution" (InD) and "out-of-distribution" (OOD) sets. Each set contains different sessions / animals. The InD data contain 789 labeled frames, and 17 unlabeled videos; the OOD data contain 253 labeled frames, and 5 unlabeled videos.
Data collected by Richard Warren and Qianyun Zhang in the Sawtell Lab, Columbia University.