Research map / Video / Video generation
Taming I2V models for Image HOI Editing: A Cognitive Benchmark and Agentic Self-Correcting Framework
Jiayi Gao, Qingchao Chen, Yuxin Peng, Yang Liu
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 SCPE (agentic playbook-driven prompt refinement) to Wan 2.2 I2V | Wan 2.2 I2V 14B · HOI-Edit L1 | Interaction (I): 0.6908 -> 0.8423 | Improved |
| add SCPE to Wan 2.2 I2V | Wan 2.2 I2V 14B · HOI-Edit L3 | I+Q&A: 0.5343 -> 0.6528 | Improved |
| use I2V model (Wan 2.2) instead of image editing models | HOI editing · HOI-Edit L1 | Interaction (I): 0.6326 (Bagel) -> 0.6908 (Wan 2.2) | Improved |
| add SCPE playbook (full) versus analyst-only variant without Playbook (ablation) | SCPE · HOI-Edit overall | Interaction (I): 0.7625 -> 0.8199 | Improved |
| add Wan official prompt enhancer (OPE) (ablation) | Wan 2.2 I2V 14B · HOI-Edit overall | Interaction (I) / IDS: I 0.6804 -> 0.7028; IDS 0.8494 -> 0.7385 | Mixed |
| add analyst visual feedback without Playbook (ablation) | Wan 2.2 I2V 14B · HOI-Edit overall | Interaction (I): 0.6804 -> 0.7625 | Improved |
| add SCPE to Nano Banana | Nano Banana · HOI-Edit overall | Interaction (I): 0.7231 -> 0.7310 | Improved |
| add SCPE to TurboDiffusion | TurboDiffusion · HOI-Edit overall | Interaction (I): 0.6290 -> 0.7536 | Improved |
| replace Gemini agent with open-source Qwen 3.5-VL in SCPE (ablation) | SCPE · HOI-Edit L1 | Interaction (I): 0.8423 -> 0.8368 | No clear effect |
| use Gemini only as prompt enhancer instead of SCPE (ablation) | Wan 2.2 I2V 14B · HOI-Edit L1 | Human consistency (H): 0.9166 -> 0.8287 | Hurt |
| increase SCPE iterations beyond R1 (ablation) | SCPE · HOI-Edit | Interaction (I) | No clear effect |
| update playbook only from failure cases instead of all cases (ablation) | SCPE · HOI-Edit | Interaction (I) | Improved |
Closest papers by meaning
- HOI-Swap: Swapping Objects in Videos with Hand-Object Interaction Awareness (NeurIPS 2024 · similarity 0.88)
- Learning Action and Reasoning-Centric Image Editing from Videos and Simulation (NeurIPS 2024 · similarity 0.88)
- I2EBench: A Comprehensive Benchmark for Instruction-based Image Editing (NeurIPS 2024 · similarity 0.88)
- Semantic-Aware Human Object Interaction Image Generation (ICML 2024 · similarity 0.87)
- HOIGen-1M: A Large-scale Dataset for Human-Object Interaction Video Generation (CVPR 2025 · similarity 0.86)
- EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editing (ICLR 2026 · 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.
