Research map / Video / Video generation
VINCIE: Unlocking In-context Image Editing from Video
Leigang Qu, Feng Cheng, Ziyan Yang, Qi Zhao, Shanchuan Lin, Yichun Shi, Yicong Li, Wenjie Wang, Tat-Seng Chua, Lu Jiang
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 segmentation prediction (CS -> I at inference) (ablation) | VINCIE (intermediate ckpt) · MSE-Bench | turn-5 success rate: 0.113 -> 0.173 | Improved |
| add segmentation tasks in training, image-only inference (ablation) | VINCIE (intermediate ckpt) · MSE-Bench | turn-5 success rate: 0.113 -> 0.103 | No clear effect |
| add segmentation (CS -> NS -> I) (ablation) | VINCIE (intermediate ckpt) · MagicBrush | turn-3 DINO: 0.592 -> 0.679 | Improved |
| train on video sequence data instead of pairwise data (ablation) | VINCIE · MSE-Bench | turn-5 success rate: 0.010 -> 0.220 | Improved |
| SFT on pairwise data after sequence pretraining (ablation) | VINCIE · MSE-Bench | turn-5 success rate: 0.220 -> 0.250 | Improved |
| add SFT to 7B model | VINCIE 7B · MSE-Bench | turn-5 success rate: 0.350 -> 0.487 | Improved |
| scale model from 3B to 7B (no SFT) | VINCIE · MSE-Bench | turn-5 success rate: 0.210 -> 0.350 | Improved |
| scale model from 3B to 7B (no SFT) | VINCIE · MSE-Bench | turn-1 success rate: 0.913 -> 0.837 | Hurt |
| add dummy context before turn-1 (ablation) | VINCIE · MagicBrush | turn-1 L1: 0.155 -> 0.086 | Improved |
| add dummy context at turn 2-3 when history present (ablation) | VINCIE · MagicBrush | turn-2 L1: 0.099 -> 0.087 | No clear effect |
| interleaved RoPE instead of text-then-image (full attention) (ablation) | VINCIE · MSE-Bench | turn-5 success rate: 0.160 -> 0.183 | Mixed |
| block-causal instead of full attention (ablation) | VINCIE · MSE-Bench | turn-5 success rate: 0.183 -> 0.120 | Hurt |
Closest papers by meaning
- VINCIE-NExT: Unlocking Video Editing from Images via In-Context Modeling (NeurIPS 2026 · similarity 0.90)
- ImgEdit: A Unified Image Editing Dataset and Benchmark (NeurIPS 2025 · similarity 0.88)
- VicEdit: Learning to Edit Videos from Visual In-Context Examples (NeurIPS 2026 · similarity 0.88)
- Zero-shot Image Editing with Reference Imitation (NeurIPS 2024 · similarity 0.87)
- Edit-R2: Context-Aware Reinforcement Learning for Multi-Turn Image Editing (NeurIPS 2026 · similarity 0.87)
- Learning Action and Reasoning-Centric Image Editing from Videos and Simulation (NeurIPS 2024 · 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.
