Research map / Detection and segmentation / Open-set and few-shot detection
FORB: A Flat Object Retrieval Benchmark for Universal Image Embedding
Pengxiang Wu, Siman Wang, Kevin S Dela Rosa, Derek Hao Hu
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 reranking with local features (ablation) | BoW · FORB overall | mAP@5 (%): 78.44 -> 80.38 | Improved |
| Add reranking with local features (ablation) | BoW · FORB overall | t-mAP@5 (%): 62.49 -> 67.83 | Improved |
| Add reranking with local features (ablation) | DELG · FORB overall | mAP@5 (%): 48.81 -> 58.74 | Improved |
| Add reranking with local features (ablation) | DELG · FORB overall | inference seconds per query: 0.376 -> 6.015 | Hurt |
| Use handcrafted BoW instead of learned DELG | retrieval embedding · FORB overall | mAP@5 (%): 48.81 -> 78.44 | Improved |
| Use mid-level FIRe features instead of BoW low-level features | retrieval embedding · FORB overall | t-mAP@5 (%): 62.49 -> 77.50 | Improved |
| Use FIRe instead of CLIP | retrieval embedding · FORB overall | mAP@5 (%): 89.36 -> 88.08 | No clear effect |
| Use FIRe instead of CLIP | retrieval embedding · FORB overall | t-mAP@5 (%): 67.23 -> 77.50 | Improved |
| Use FIRe instead of CLIP for speed | retrieval embedding · FORB inference | seconds per query: 0.513 -> 0.124 | Improved |
| Larger model and training data for top-only methods (CLIP vs SLIP) | global embedding · FORB overall | mAP@5 (%): 39.01 -> 89.36 | Improved |
| Larger training data (DiHT vs DINO) | global embedding · FORB overall | mAP@5 (%): 55.20 -> 84.77 | Improved |
| Use top-only CLIP instead of FIRe on logos / hard split | retrieval embedding · FORB hard split | mAP@5 (%): 56.58 -> 73.84 | Improved |
Closest papers by meaning
- Detect-To-Retrieve: Efficient Regional Aggregation for Image Search (CVPR 2019 · similarity 0.88)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.87)
- Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval (ICLR 2025 · similarity 0.87)
- Dual Pose-invariant Embeddings: Learning Category and Object-specific Discriminative Representations for Recognition and Retrieval (CVPR 2024 · similarity 0.87)
- On Train-Test Class Overlap and Detection for Image Retrieval (CVPR 2024 · similarity 0.87)
- Search and Detect: Training-Free Long Tail Object Detection via Web-Image Retrieval (CVPR 2025 · 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.
