Research map / Detection and segmentation / CNN features
Enhanced Training of Query-Based Object Detection via Selective Query Recollection
Fangyi Chen, Han Zhang, Kai Hu, Yu-Kai Huang, Chenchen Zhu, Marios Savvides
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 |
|---|---|---|---|
| Apply SQR | DAB-DETR-R50 · COCO 2017 val | AP: 42.2 -> 44.5 (+2.3) | Improved |
| Apply SQR | DAB-DETR-SwinB · COCO 2017 val | AP: 49.0 -> 51.6 (+2.6) | Improved |
| Apply SQR | Deformable DETR-R50 · COCO 2017 val, 12 epochs | AP: 37.2 -> 39.9 (+2.7) | Improved |
| Apply SQR | Deformable DETR-R50 · COCO 2017 val, 50 epochs | AP: 44.5 -> 45.9 (+1.4) | Improved |
| Apply SQR | DN-DETR · COCO 2017 val, 12 epochs | AP: 38.5 -> 40.4 (+1.9) | Improved |
| Apply SQR | DN-DETR · COCO 2017 val, 50 epochs | AP: 44.1 -> 45.2 (+1.1) | Improved |
| Replace selective recollection with dense query recollection (ablation) | Adamixer-R50 · COCO val, 1x schedule | AP: 42.5 -> 44.2 (+1.7 AP) | Improved |
| Selectively recollect queries from the prior two stages rather than recollecting all prior queries (ablation) | Adamixer-R50 · COCO val, 1x schedule | AP: 44.2 -> 44.4 | Improved |
| Start query recollection at stage 2 instead of stage 1 (ablation) | SQR on Adamixer-R50 · COCO val | AP: 44.4 -> 44.2 | Hurt |
| Start query recollection at stage 3 instead of stage 1 (ablation) | SQR on Adamixer-R50 · COCO val | AP: 44.4 -> 43.8 | Hurt |
| Start query recollection at stage 4 instead of stage 1 (ablation) | SQR on Adamixer-R50 · COCO val | AP: 44.4 -> 42.9 | Hurt |
| Reweight stage losses in Fibonacci sequence without query recollection (ablation) | Adamixer-R50 · COCO val | AP: -0.6 | Hurt |
Closest papers by meaning
- Order Matters: Competition-Guided Query Ordering for RNN-Based Object Detection (NeurIPS 2026 · similarity 0.88)
- EASE-DETR: Easing the Competition among Object Queries (CVPR 2024 · similarity 0.87)
- CQ-DINO: Mitigating Gradient Dilution via Category Queries for Vast Vocabulary Object Detection (NeurIPS 2025 · similarity 0.86)
- Salience DETR: Enhancing Detection Transformer with Hierarchical Salience Filtering Refinement (CVPR 2024 · similarity 0.85)
- MI-DETR: An Object Detection Model with Multi-time Inquiries Mechanism (CVPR 2025 · similarity 0.85)
- MS-DETR: Efficient DETR Training with Mixed Supervision (CVPR 2024 · similarity 0.85)
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.
