Research map / Graph learning / Graph self-supervised learning
Semantic Derivative Flow: Graph-Guided Diffusion for Controllable Instance Interactions
Shibin Mei, Hang Wang, Bingbing Ni
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 derivative attention instead of independent subject/predicate/object encoding (ablation) | SDF (SD1.5) · HICODet Swin-Large Default Full | HOI detection mAP: 31.87 -> 32.20 | Improved |
| add global context node (ablation) | SDF (SD1.5) · HICODet Swin-Large Default Full | HOI detection mAP: 32.08 -> 32.20 | Improved |
| add predicate box expansion (ablation) | SDF (SD1.5) · HICODet Swin-Large Default Full | HOI detection mAP: 32.16 -> 32.20 | Improved |
| add regional refinement module (ablation) | SDF (SD1.5) · HICODet Swin-Large Default Full | HOI detection mAP: 32.02 -> 32.20 | Improved |
| add derivative attention (ablation) | SDF (SD1.5) · HICODet | FID: 18.65 -> 18.58 | Improved |
| set predicate box expansion zeta to 0.1 vs 0.00 (ablation) | SDF (SD1.5) · HICODet Default Full (Swin-Tiny) | HOI detection mAP: 29.10 -> 30.14 | Improved |
| increase expansion zeta from 0.1 to 0.2 (ablation) | SDF (SD1.5) · HICODet Default Full (Swin-Tiny) | HOI detection mAP: 30.14 -> 30.11 | No clear effect |
| increase ROIAlign output size from 1x1 to 3x3 (ablation) | SDF (SD1.5) · HICODet Default Full (Swin-Tiny) | HOI detection mAP: 30.14 -> 30.13 | No clear effect |
| use SDF instead of InteractDiffusion (SD1.5) | interaction-controlled generation · HICODet Default Rare (Swin-Tiny) | HOI detection mAP: 23.02 -> 23.85 | Improved |
| use SDXL backbone instead of SD1.5 for SDF | SDF · HICODet Default Full (Swin-Tiny) | HOI detection mAP: 30.14 -> 31.37 | Improved |
| use SDF instead of InteractDiffusion | interaction-controlled generation · HICODet | FID: 18.69 -> 18.58 | Improved |
Closest papers by meaning
- R3CD: Scene Graph to Image Generation with Relation-Aware Compositional Contrastive Control Diffusion (AAAI 2024 · similarity 0.87)
- InteractDiffusion: Interaction Control in Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.87)
- VerbDiff: Text-Only Diffusion Models with Enhanced Interaction Awareness (CVPR 2025 · similarity 0.87)
- SSMG: Spatial-Semantic Map Guided Diffusion Model for Free-Form Layout-to-Image Generation (AAAI 2024 · similarity 0.86)
- The Hidden Language of Diffusion Models (ICLR 2024 · similarity 0.86)
- Ranni: Taming Text-to-Image Diffusion for Accurate Instruction Following (CVPR 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.
