Research map / Video / Video generation
Goku: A Million-Scale Universal Dataset and Benchmark for Instruction-Based Video Editing
Sen Liang, Cong Wang, Zhentao Yu, Fengbin Guan, zhengguang zhou, Teng Hu, youliang zhang, Yuan Zhou, Xin Li, Qinglin Lu, Zhibo Chen
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 |
|---|---|---|---|
| compare Goku-Edit with best open-source LucyEdit | Goku-Edit · Goku-Bench instruction task | IF: 0.549 -> 0.627 | Improved |
| compare Goku-Edit with closed-source Runway | Goku-Edit · Goku-Bench instruction task | IF: 0.758 (Runway) vs 0.627 | Hurt |
| Goku-Edit vs Runway on physical and camera metrics | Goku-Edit · Goku-Bench | PR / CM: PR 0.705 -> 0.738; CM 0.891 -> 0.927 | Improved |
| use MLLM text encoder (ablation) | Goku-Edit · Goku-Bench | IF: 0.541 -> 0.573 | Improved |
| add dual-branch mask architecture (ablation) | Goku-Edit · Goku-Bench | SR: 0.756 -> 0.789 | Improved |
| add RoPE-aligned spatial cross-attention (ablation) | Goku-Edit · Goku-Bench | SR: 0.789 -> 0.819 | Improved |
| add SpatialCFG at inference (ablation) | Goku-Edit · Goku-Bench | PR: 0.718 -> 0.738 | Improved |
| train on unfiltered vs progressively filtered Goku (ablation) | LucyEdit architecture · Goku-Bench multi-task | IF / EQ | Improved |
| scale training data from 50k to 100k (ablation) | LucyEdit architecture · Goku-Bench multi-task | all metrics | Improved |
| spatial downsampling factor n=8 instead of 4 (ablation) | Goku-Edit · qualitative | localization | Hurt |
Closest papers by meaning
- Señorita-2M: A High-Quality Instruction-based Dataset for General Video Editing by Video Specialists (NeurIPS 2025 · similarity 0.89)
- AnyEdit: Mastering Unified High-Quality Image Editing for Any Idea (CVPR 2025 · similarity 0.88)
- VINCIE-NExT: Unlocking Video Editing from Images via In-Context Modeling (NeurIPS 2026 · similarity 0.88)
- InsightEdit: Towards Better Instruction Following for Image Editing (CVPR 2025 · similarity 0.87)
- Learning Action and Reasoning-Centric Image Editing from Videos and Simulation (NeurIPS 2024 · similarity 0.87)
- Instruction-based Image Manipulation by Watching How Things Move (CVPR 2025 · similarity 0.87)
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.
