CoolFace
Apppublic

Rongjin03/spatialvid-trajectory-clusters

sourceHugging Facecc-by-nc-sa-4.0updated 5d agoView on Hugging Face
0likes
App README

SpatialVID Trajectory Clusters

Interactive review of 5,000 trajectories from SpatialVID group_0001 at several K values. For K=8, 16, 32, 48, and 64, the Space shows up to three review samples per geometric cluster: a center-near prototype, a mid-cluster sample, and a moderately outer sample. Each sample pairs the video with a shape-normalized camera trajectory reconstructed from the released poses. The trajectory panel supports an interactive 3D view and X-Z, Z-Y, and X-Y orthographic projections. The three representative trajectories are also overlaid after aligning their first camera pose and normalizing path length. Short line segments show camera viewing direction. The Play all control starts the three visible samples together with audio muted.

The clusters describe pose geometry, not scene semantics or capture hardware. The selected clips are not guaranteed to be drone footage, and the visualization does not validate pose accuracy. Video and annotations originate from SpatialVID and remain under CC BY-NC-SA 4.0. The selected media files are hosted in the companion media repository.