Research map / Image generation / Text-to-image diffusion
DiffSim: Taming Diffusion Models for Evaluating Visual Similarity
Yiren Song, Xiaokang Liu, Mike Zheng Shou
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 Aligned Attention Score alignment to diffusion features (versus raw cosine) (ablation) | DiffSim · Sref bench | accuracy: 78.80% -> 97.40% | Improved |
| add Aligned Attention Score alignment to diffusion features (versus raw cosine) (ablation) | DiffSim · IP bench | accuracy: 62.47% -> 92.04% | Improved |
| add Aligned Attention Score alignment to diffusion features (versus raw cosine) (ablation) | DiffSim · NIGHTS | accuracy: 66.75% -> 86.82% | Improved |
| add AAS to CLIP features (ablation) | CLIP · Sref bench | accuracy: 66.50% -> 71.15% | Improved |
| add AAS to CLIP features (ablation) | CLIP · IP bench | accuracy: 82.54% -> 87.36% | Improved |
| add AAS to DINO v2 features (ablation) | DINO v2 · NIGHTS | accuracy: 81.00% -> 86.41% | Improved |
| use cross-attention (DiffSim-C) instead of self-attention (DiffSim-S) (ablation) | DiffSim · IP bench | accuracy: 92.04% -> 77.06% | Hurt |
| use cross-attention (DiffSim-C) instead of self-attention (DiffSim-S) (ablation) | DiffSim · CUTE | accuracy: 72.06% -> 76.17% | Improved |
| use SD-XL backbone instead of SD1.5 (ablation) | DiffSim-S · NIGHTS | accuracy: 86.52% -> 78.05% | Hurt |
| increase input resolution from 512 to 768 (ablation) | DiffSim · IP bench and Sref bench | accuracy | No clear effect |
| use higher denoising timestep (900) versus lower (500) (ablation) | DiffSim · Sref (style) versus NIGHTS (instance) | accuracy | Mixed |
| ensemble CLIP, DINO v2 and DiffSim by hard vote | DiffSim · NIGHTS | accuracy: 86.52% -> 89.43% | Improved |
Closest papers by meaning
- Diff-2-in-1: Bridging Generation and Dense Perception with Diffusion Models (ICLR 2025 · similarity 0.88)
- Diff-Plugin: Revitalizing Details for Diffusion-based Low-level Tasks (CVPR 2024 · similarity 0.86)
- The Many Senses of Visual Similarity: A Text-Prompted Image Perceptual Metric (NeurIPS 2026 · similarity 0.86)
- Diffusion Model for Dense Matching (ICLR 2024 · similarity 0.85)
- DiffBEV: Conditional Diffusion Model for Bird’s Eye View Perception (AAAI 2024 · similarity 0.85)
- Aesthetic Post-Training Diffusion Models from Generic Preferences with Step-by-step Preference Optimization (CVPR 2025 · similarity 0.85)
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.
