Research map / Video / Video segmentation and tracking
BeyondMasks: Evaluating Causal and Physical Consistency in Video Object Removal
Yiğit Ekin, Enes Şanlı, Aykut Erdem, Erkut Erdem, Aysegul Dundar
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 CORE (Gemini) instead of LPIPS as the metric | evaluation of video object removal · BeyondMasks human alignment | Spearman correlation with human ObjectScore: -0.02 -> 0.62 | Improved |
| use CORE (Gemini) instead of PSNR as the metric | evaluation of video object removal · BeyondMasks human alignment | Spearman correlation with human AfterEffectScore: 0.28 -> 0.73 | Improved |
| swap judge VLM from Gemini to GPT (ablation) | CORE · BeyondMasks human alignment | Spearman correlation (ObjectScore / AfterEffectScore): 0.62 / 0.73 -> 0.60 / 0.76 | No clear effect |
| rank by CORE-AES instead of PSNR (DiffuEraser vs Generative Omnimatte) | video object removal methods · BeyondMasks | CORE-AES: PSNR 25.1880 vs 23.8106; CORE-AES 2.625 vs 3.021 | Mixed |
| move from synthetic to real subset | ROSE · BeyondMasks | CORE-OS: 4.068 -> 3.505 | Hurt |
| use frame-wise ObjectClear instead of best video method on synthetic subset | ROSE · BeyondMasks (synthetic) | CORE-OS: 4.068 -> 4.325 | Improved |
| use mask-guided ProPainter instead of text-only Lucy Edit | video object removal · BeyondMasks | CORE-OS: 1.443 -> 3.063 | Improved |
| evaluate steam instead of shadow after-effects | top six video removal methods · BeyondMasks | CORE-AES: 2.520 -> 2.032 | Hurt |
Closest papers by meaning
- VOID: Video Object and Interaction Deletion (ECCV 2026 · similarity 0.92)
- EffectErase: Joint Video Object Removal and Insertion for High-Quality Effect Erasing (CVPR 2026 · similarity 0.91)
- GeoRemover: Removing Objects and Their Causal Visual Artifacts (NeurIPS 2025 · similarity 0.87)
- ROSE: Remove Objects with Side Effects in Videos (NeurIPS 2025 · similarity 0.87)
- Generative Omnimatte: Learning to Decompose Video into Layers (CVPR 2025 · similarity 0.87)
- MiniMax-Remover: Taming Bad Noise Helps Video Object Removal (NeurIPS 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.
