Research map / Video / Video generation
Layer-Aware Video Composition via Split-then-Merge
Ozgur Kara, Yujia Chen, Ming-Hsuan Yang, James Rehg, Wen-Sheng Chu, Du Tran
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 |
|---|---|---|---|
| add transformation-aware foreground augmentation (ablation) | StM · StM test | M1/M3 (FG): FG ID 83.15 -> 84.82; BG motion 16.61 vs 16.36 | Mixed |
| add identity-preservation loss (ablation) | StM · StM test | M4 BG motion: 23.75 -> 16.36 | Improved |
| add identity-preservation loss (ablation) | StM · StM test | M2 BG identity: 88.25 -> 92.88 | Improved |
| add both augmentation and ID loss (vs neither) (ablation) | StM · StM test | M2 BG identity: 84.76 -> 92.88 | Improved |
| add both components (vs neither) (ablation) | StM · StM test | M1 FG identity: 90.39 -> 84.82 | Hurt |
| scale training data 10K -> 50K clips (ablation) | StM · StM test | M1 FG identity: 77.96 -> 84.82 | Improved |
| scale training data 10K -> 50K clips (ablation) | StM · StM test | M4 BG motion: 22.27 -> 16.36 | Improved |
| increase FG mask random pixel masking to 50% (ablation) | StM · SAM-3 eval | M1: 82.22 -> 73.27 | Hurt |
| use full video layers instead of static image conditioning | StM vs Qwen + I2V · StM test | M4 BG motion: 74.77 -> 16.36 | Improved |
Closest papers by meaning
- CAGE: Unsupervised Visual Composition and Animation for Controllable Video Generation (AAAI 2025 · similarity 0.87)
- TTVidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining (NeurIPS 2026 · similarity 0.87)
- Generative Omnimatte: Learning to Decompose Video into Layers (CVPR 2025 · similarity 0.87)
- Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance (NeurIPS 2025 · similarity 0.87)
- ONE-SHOT: Compositional Human-Environment Video Synthesis via Spatial-Decoupled Motion Injection and Hybrid Context Integration (NeurIPS 2026 · similarity 0.86)
- VideoJAM: Joint Appearance-Motion Representations for Enhanced Motion Generation in Video Models (ICML 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.
