JavisDiT audio-visual object removal

@ — 10 held-out clips from test.jsonl, each shown as input / generated / target / pseudo GT.

What the four videos are.

input — VAE decode of the stored source latent: the clip with the object still present.
generated — the model's edit for the prompt a video without {object}.
target — VAE decode of the stored target latent, i.e. the exact tensor the training loss used.
pseudo GT — the raw artifacts that target latent was encoded from: the inpainted video muxed with the SAM-Audio residual track. Pseudo because no real "object removed" recording exists — the pair is synthesised by an inpainting model and a source-separation model.

Target and pseudo GT are the same content either side of the VAE round trip, so the L1 between them (VAE floor) is a lower bound no generation can beat.

Per-clip metrics

Mean absolute pixel difference, 0–255. gap = L1→target − L1→input; negative means the edit moved toward the target rather than copying its input.

256×256, 81 frames @ 16 fps (5.06 s), mono 16 kHz. cfg , sampling steps, no prepended reference frame. Per-entry seeds (cfg.seed + crc32(video_id)).