Research map / Image generation / Fast diffusion sampling
Ensemble Diffusion for Retrieval
Song Bai, Zhichao Zhou, Jingdong Wang, Xiang Bai, Longin Jan Latecki, Qi Tian
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 learned sparse weights (RED) instead of equal-weight naive fusion (ablation) | diffusion-based fusion · ModelNet40 | mAP: 84.55 -> 86.30 | Improved |
| add learned sparse weights (RED) instead of equal-weight naive fusion (ablation) | diffusion-based fusion · ModelNet40 | AUC: 85.26 -> 87.03 | Improved |
| add learned weights (RED) instead of equal-weight naive fusion (ablation) | diffusion-based fusion · Holidays | mAP: 90.69 -> 93.32 | Improved |
| add learned weights (RED) instead of equal-weight naive fusion (ablation) | diffusion-based fusion · Ukbench | N-S score: 3.907 -> 3.938 | Improved |
| replace the exponential weight learner of SN with RED's sparse-capable weight learning (ablation) | similarity fusion · Holidays | mAP: 91.72 -> 93.32 | Improved |
| fuse the weak PANORAMA similarity with Vol. CNN using tensor product fusion (ablation) | Vol. CNN · ModelNet40 | mAP: 79.53 -> 67.16 | Hurt |
| fuse GIFT with Vol. CNN using tensor product fusion (ablation) | Vol. CNN · ModelNet40 | mAP: 79.53 -> 83.23 | Improved |
| use RED instead of the best tensor product fusion pair (ablation) | similarity fusion · ModelNet40 | mAP: 85.12 -> 86.30 | Improved |
| fuse 5 random noisy similarities with naive fusion (ablation) | naive fusion · Holidays | mAP: 14.82 | Hurt |
| fuse 5 random noisy similarities with RED (ablation) | RED · Holidays | mAP | No clear effect |
| vary the weight regularizer lambda in RED (ablation) | RED · Holidays | mAP | No clear effect |
| swap the order of fusion (GIFT with ResNet) in tensor product fusion (ablation) | tensor product fusion · ModelNet40 | mAP: 85.08 vs 85.12 | Mixed |
Closest papers by meaning
- Cluster-Aware Similarity Diffusion for Instance Retrieval (ICML 2024 · similarity 0.85)
- Where's Waldo: Diffusion Features For Personalized Segmentation and Retrieval (NeurIPS 2024 · similarity 0.84)
- Mining your own secrets: Diffusion Classifier Scores for Continual Personalization of Text-to-Image Diffusion Models (ICLR 2025 · similarity 0.84)
- Diffusion Classifiers Understand Compositionality, but Conditions Apply (NeurIPS 2025 · similarity 0.83)
- Diffusion Model for Dense Matching (ICLR 2024 · similarity 0.83)
- Ensembling Diffusion Models via Adaptive Feature Aggregation (ICLR 2025 · similarity 0.83)
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.
