Research map / Video / Video generation
EasyOmnimatte: Taming Pretrained Inpainting Diffusion Models for End-to-End Video Layered Decompositio
Yihan Hu, Xuelin Chen, Xiaodong Cun
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 |
|---|---|---|---|
| replace multi-stage optimization (Gen-Omnimatte) with end-to-end dual-expert model | EasyOmnimatte · Video omnimatte (reconstruction/composition) | FVD: 116.32 -> 105.48 | Improved |
| end-to-end model vs Gen-Omnimatte | EasyOmnimatte · User study (0-5) | Overall score: 2.85 -> 4.08 | Improved |
| EasyOmnimatte vs BGM V2 on reconstruction PSNR | EasyOmnimatte · Video omnimatte (reconstruction/composition) | PSNR: 26.61 -> 26.23 | No clear effect |
| apply LoRA to all DiT blocks instead of only effect-sensitive blocks (ablation) | EasyOmnimatte · Video omnimatte | effect capture | Hurt |
| add Effect Expert with LoRA only on final-stage blocks (ablation) | EasyOmnimatte · Video omnimatte | effect preservation | Improved |
| Effect Expert only (no Quality Expert) (ablation) | EasyOmnimatte · Video omnimatte | matte boundary precision | Hurt |
| Quality Expert only (no Effect Expert) (ablation) | EasyOmnimatte · Video omnimatte | effect capture | Hurt |
| combine Effect and Quality experts via timestep threshold tau=0.5 (ablation) | EasyOmnimatte · Video omnimatte | matte quality and effects | Improved |
| allow unrestricted attention between inpainting and matting tokens at inference (ablation) | EasyOmnimatte · Video omnimatte | background/foreground decomposition | Hurt |
| fine-tune general video model (WAN Fun) instead of inpainting model (ablation) | EasyOmnimatte · Video omnimatte | effect capture | Hurt |
Closest papers by meaning
- Generative Omnimatte: Learning to Decompose Video into Layers (CVPR 2025 · similarity 0.92)
- Prompt2Effect: Training-Free LoRA Synthesis for Controllable Video Effects (ECCV 2026 · similarity 0.87)
- HomoGen: Enhanced Video Inpainting via Homography Propagation and Diffusion (CVPR 2025 · similarity 0.86)
- Emergent Temporal Correspondences from Video Diffusion Transformers (NeurIPS 2025 · similarity 0.86)
- Motion Consistency Model: Accelerating Video Diffusion with Disentangled Motion-Appearance Distillation (NeurIPS 2024 · similarity 0.86)
- Video Diffusion Models Are Strong Video Inpainter (AAAI 2025 · similarity 0.86)
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.
