Research map / Detection and segmentation / CNN features
Anchor DETR: Query Design for Transformer-Based Detector
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 anchor-point queries and multiple patterns per anchor (ablation) | Anchor DETR · MS COCO val2017 | AP: 39.3 to 44.2 | Helped |
| add anchor-point queries (ablation) | Anchor DETR with standard attention · MS COCO val2017 | AP: 39.3 to 42.6 | Helped |
| add multiple patterns to each anchor point (ablation) | Anchor-point queries · MS COCO val2017 | AP: 42.6 to 44.2 | Helped |
| increase patterns from 1 to 3 per anchor (ablation) | Anchor DETR with 100 anchor points · MS COCO val2017 | AP: 40.8 to 43.4 | Helped |
| increase anchor points from 300 to 900 with one pattern (ablation) | Anchor DETR · MS COCO val2017 | AP: 42.6 to 42.9 | No clear effect |
| use grid instead of learned anchor points (ablation) | Anchor DETR · MS COCO val2017 | AP: 44.2 to 44.1 | No clear effect |
| replace standard attention with RCDA (ablation) | Anchor DETR with DC5 features · MS COCO val2017 | AP: 44.2 to 44.2 | No clear effect |
| replace standard attention with RCDA (ablation) | Anchor DETR with DC5 features · MS COCO val2017 | training memory: 10.5G to 4.4G | Helped |
| replace standard attention with RCDA (ablation) | Anchor DETR with C5 features · MS COCO val2017 | AP: 42.2 to 42.1 | No clear effect |
| use Anchor DETR instead of DETR with the same DC5 feature | ResNet-50 detector · MS COCO val2017 | AP: 43.3 to 44.2 | Helped |
| use Anchor DETR instead of DETR with the same DC5 feature | ResNet-50 detector · MS COCO val2017 | FPS: 12 to 19 | Helped |
Closest papers by meaning
- SAP-DETR: Bridging the Gap Between Salient Points and Queries-Based Transformer Detector for Fast Model Convergency (CVPR 2023 · similarity 0.90)
- FP-DETR: Detection Transformer Advanced by Fully Pre-training (ICLR 2022 · similarity 0.90)
- UP-DETR: Unsupervised Pre-Training for Object Detection With Transformers (CVPR 2021 · similarity 0.89)
- KD-DETR: Knowledge Distillation for Detection Transformer with Consistent Distillation Points Sampling (CVPR 2024 · similarity 0.88)
- DECO: Unleashing the Potential of ConvNets for Query-based Detection and Segmentation (ICLR 2025 · similarity 0.87)
- Mr. DETR: Instructive Multi-Route Training for Detection Transformers (CVPR 2025 · similarity 0.87)
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.