Research map / Detection and segmentation / Open-set and few-shot detection
Efficient Discovery and Effective Evaluation of Visual Perceptual Similarity: A Benchmark and Beyond
Oren Barkan, Tal Reiss, Jonathan Weill, Ori Katz, Roy Hirsch, Itzik Malkiel, Noam Koenigstein
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 |
|---|---|---|---|
| label model-suggested top-k suspects (EDS) instead of random pairs | visual similarity labeling · DeepFashion in-catalog | positive discovery rate: 0.001 -> 0.848 | Improved |
| finetune with identity labels (ablation) | Argus · Closed-catalog discovery | Micro ROC-AUC: 62.7 -> 62.3 | Hurt |
| finetune with identity labels (ablation) | Argus · Closed-catalog discovery | HR@5: 82.5 -> 84.3 | Improved |
| finetune with category labels (ablation) | BEiT · Closed-catalog discovery | Micro ROC-AUC: 75.5 -> 72.9 | Hurt |
| finetune with category labels (ablation) | Argus · Closed-catalog discovery | Micro ROC-AUC: 62.7 -> 67.5 | Improved |
| supervised finetuning (CAT, ID, ID+C) (ablation) | DINO · Image-in-the-wild discovery | Micro ROC-AUC: 70.8 -> 64.8 / 61.2 / 62.0 | Hurt |
| finetune with DINO's own self-supervised objective (ablation) | DINO · Image-in-the-wild discovery | Micro ROC-AUC: 70.8 -> 71.8 | Improved |
| evaluate with HR@k/MRR@k on non-generator models | benchmark metrics · Closed-catalog discovery | HR@5: 92.3 (generator BEiT) vs 38.7 (BEiT CAT) | Hurt |
| use ROC-AUC with leave-one-out ground truth subsets (ablation) | evaluation metric · Closed-catalog discovery | Spearman correlation: 0.94 | Improved |
Closest papers by meaning
- The Many Senses of Visual Similarity: A Text-Prompted Image Perceptual Metric (NeurIPS 2026 · similarity 0.92)
- Interpretable Measures of Conceptual Similarity by Complexity-Constrained Descriptive Auto-Encoding (CVPR 2024 · similarity 0.89)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.87)
- Same or Not? Enhancing Visual Perception in Vision-Language Models (CVPR 2026 · similarity 0.87)
- Visual Data-Type Understanding does not emerge from scaling Vision-Language Models (ICLR 2024 · similarity 0.86)
- Image2Struct: Benchmarking Structure Extraction for Vision-Language Models (NeurIPS 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.
