Research map / Detection and segmentation / Open-set and few-shot detection
The Pursuit of Knowledge: Discovering and Localizing Novel Categories Using Dual Memory
Sai Saketh Rambhatla, Rama Chellappa, Abhinav Shrivastava
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 |
|---|---|---|---|
| initialize semantic memory with known-class detector scores (ablation) | Dual memory discovery · COCO mini-train (5000 imgs), semantic memory init ablation | AuC@0.5: 2.31 -> 2.97 | Improved |
| initialize semantic memory with ground-truth overlap centroids instead of detector scores (ablation) | Dual memory discovery · COCO mini-train (5000 imgs), semantic memory init ablation | # discovered objects: 27 -> 30 | Improved |
| add merge step to memory consolidation (ablation) | Dual memory discovery · COCO mini-train (5000 imgs), consolidation ablation | AuC@0.5: 3.70 -> 3.80 | Improved |
| add refine step after merge (ablation) | Dual memory discovery · COCO mini-train (5000 imgs), consolidation ablation | AuC@0.5: 3.80 -> 4.02 | Improved |
| add refine step after merge (ablation) | Dual memory discovery · COCO mini-train (5000 imgs), consolidation ablation | CorLoc: 42.57 -> 41.20 | Hurt |
| add refine step after merge (ablation) | Dual memory discovery · COCO mini-train (5000 imgs), consolidation ablation | CorRet: 62.21 -> 65.88 | Improved |
| compare to K-means on full COCO train2014 | Dual memory discovery · COCO train2014 full (80k imgs), unknown classes | AuC@0.2: 7.23 -> 9.11 | Improved |
| compare to FINCH on full COCO train2014 | Dual memory discovery · COCO train2014 full (80k imgs), unknown classes | # discovered objects: 42 -> 46 | Improved |
| compare to OSD at 2.5k images | Dual memory discovery · COCO train2014 subset 2.5k imgs | CorLoc: 6.62 -> 43.00 | Improved |
| compare to OSD at 2.5k images | Dual memory discovery · COCO train2014 subset 2.5k imgs | CorRet: 80.00 -> 64.22 | Hurt |
| replace OSD proposals and features with Faster R-CNN ones | OSD · COCO train2014 subset 2.5k imgs | CorLoc: 6.62 -> 6.34 | Hurt |
| compare to rOSD at 20k images | Dual memory discovery · COCO train2014 subset 20k imgs | CorLoc: 15.77 -> 41.41 | Improved |
Closest papers by meaning
- Debiased Novel Category Discovering and Localization (AAAI 2024 · similarity 0.87)
- Semantic-Guided Novel Category Discovery (AAAI 2024 · similarity 0.86)
- Looking Beyond the Known: Towards a Data Discovery Guided Open-World Object Detection (NeurIPS 2025 · similarity 0.86)
- CAKE: Category Aware Knowledge Extraction for Open-Vocabulary Object Detection (AAAI 2025 · similarity 0.86)
- Cyclic Contrastive Knowledge Transfer for Open-Vocabulary Object Detection (ICLR 2025 · similarity 0.86)
- MOS: Modeling Object-Scene Associations in Generalized Category Discovery (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.
