Research map / Image generation / Fast diffusion sampling
Emergent Correspondence from Image Diffusion
Luming Tang, Menglin Jia, Qianqian Wang, Cheng Perng Phoo, Bharath Hariharan
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 |
|---|---|---|---|
| average features over a batch of noise samples (ablation) | DIFT_sd · SPair-71k per point | PCK: 57.9 -> 59.5 | Improved |
| average features over a batch of noise samples (ablation) | DIFT_adm · SPair-71k per point | PCK: 51.1 -> 52.0 | Improved |
| use class prompt 'a photo of a [class]' instead of null prompt (ablation) | DIFT_sd · SPair-71k per point | PCK: 57.6 -> 59.5 | Improved |
| use DDIM inversion instead of forward noising (ablation) | DIFT_sd · semantic correspondence | PCK | No clear effect |
| use SD's VAE encoder features instead of U-Net features (ablation) | DIFT_sd · all benchmarks | PCK | Hurt |
| use diffusion features (DIFT_adm) instead of DINO features, same ImageNet data | feature extractor · SPair-71k per image | PCK: 33.9 -> 46.3 | Improved |
| use diffusion features (DIFT_sd) instead of OpenCLIP features | feature extractor · SPair-71k per image | PCK: 38.4 -> 52.9 | Improved |
| use diffusion features (DIFT_adm) instead of DINO | feature extractor · DAVIS-2017 | J&F_m: 71.4 -> 75.7 | Improved |
| use diffusion features (DIFT_sd) instead of OpenCLIP | feature extractor · HPatches epsilon=5 | accuracy: 78.0 -> 83.1 | Improved |
| use diffusion features (DIFT_sd) instead of DINO | feature extractor · DAVIS-2017 | J&F_m: 71.4 (DINO) vs 70.0 (DIFT_sd) | Mixed |
Closest papers by meaning
- Unleashing Diffusion Transformers for Visual Correspondence by Modulating Massive Activations (NeurIPS 2025 · similarity 0.89)
- Emergent Temporal Correspondences from Video Diffusion Transformers (NeurIPS 2025 · similarity 0.88)
- Diffusion Model for Dense Matching (ICLR 2024 · similarity 0.88)
- CleanDIFT: Diffusion Features without Noise (CVPR 2025 · similarity 0.88)
- Diffusion 3D Features (Diff3F): Decorating Untextured Shapes with Distilled Semantic Features (CVPR 2024 · similarity 0.87)
- Multistep Distillation of Diffusion Models via Moment Matching (NeurIPS 2024 · 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.
