wantrack_synth_toy

20 synthetic Wan2.2-14B-I2V videos generated from a single shared first frame, with CoTracker3 tracks, FastSAM object_ids, and lowrank track_weights baked in. Used for WanTrack (MotionStream-style) sparse-track overfit experiments.

Layout

• synthetic_seed.png -- shared first frame used by all 20 I2V generations.
• videos/vid_NNNNNN.mp4 -- 20 clips, 121 frames, 480x832 @ 24 fps.
• tracks/vid_NNNNNN.npz -- CoTracker3 50x50 grid tracks (tracks, visibility) + object_ids from FastSAM + lowrank track_weights.
• sam_labels/vid_NNNNNN.npy -- FastSAM label maps at frame 0 (int16, -1 = background).
• preprocessed_i2v_track_varied/ -- Wan2.1-Fun-InP parquet preprocessing with per-clip captions.
• preprocessed_i2v_track_generic/ -- same clips, single generic caption for all.
• videos2caption*.json / merge*.txt / prompts.txt -- manifests, merge files, per-clip prompts.