Research map / Video / Video generation
TACO: Taming Diffusion for in-the-wild Video Amodal Completion
Ruijie Lu, Yixin Chen, Yu Liu, Jiaxiang Tang, Junfeng Ni, Diwen Wan, Gang Zeng, Siyuan Huang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use TACO video diffusion instead of image-level amodal completion (pix2gestalt) | TACO · OvO-Hard | FVD: 674.24 -> 291.49 | Improved |
| TACO vs Diffusion-VAS (explicit amodal mask prediction) on mask alignment | TACO · Kubric-Static | IoU: 84.3 -> 83.9 | No clear effect |
| progressive training (OvO-Easy then OvO-Hard) vs training from SVD weights (ablation) | TACO on Bdd100k · Bdd100k and OvO-Hard | quality | Improved |
| use TACO amodal completion as a drop-in module before object reconstruction | NeRF2Mesh · Compositional-GSO | Chamfer Distance: 4.39 -> 1.26 | Improved |
| add TACO amodal completion to POPE pose estimation | POPE · YCB-Video | Rot-Acc: 26.5 -> 34.7 | Improved |
| TACO zero-shot vs EoRaS trained on MOVi-D | TACO · MOVi-D amodal segmentation | mIoU: 61.3 -> 69.4 (EoRaS) | Hurt |
| use top-3 samples instead of top-1 for amodal segmentation (ablation) | TACO · MOVi-B | mIoU: 73.1 -> 77.8 | Improved |
| TACO vs baselines in user study | TACO · User study 36 participants | CQ score: 2.88 -> 4.65 (vs pix2gestalt) | Improved |
| mask E2FGVI inpainting with bounding-box constraint instead of visible-object mask (ablation) | E2FGVI baseline · baseline fairness | quality | Improved |
Closest papers by meaning
- WildCAT3D: Appearance-Aware Multi-View Diffusion in the Wild (NeurIPS 2025 · similarity 0.86)
- Using Diffusion Priors for Video Amodal Segmentation (CVPR 2025 · similarity 0.86)
- VD3D: Taming Large Video Diffusion Transformers for 3D Camera Control (ICLR 2025 · similarity 0.86)
- AC3D: Analyzing and Improving 3D Camera Control in Video Diffusion Transformers (CVPR 2025 · similarity 0.85)
- LongScape: Advancing Long-Horizon Embodied World Models with Context-Aware MoE (NeurIPS 2026 · similarity 0.85)
- ViVid-1-to-3: Novel View Synthesis with Video Diffusion Models (CVPR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
