Research map / Detection and segmentation / CNN features
DESTR: Object Detection With Split Transformer
Check 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 the mini-detector to initialize decoder queries (ablation) | C-DETR · COCO-val | AP: 40.9 -> 41.5 | Helped |
| split cross-attention into classification and regression branches (ablation) | C-DETR · COCO-val | AP: 40.9 -> 41.4 | Helped |
| add mini-detector and split cross-attention together (ablation) | C-DETR · COCO-val | AP: 40.9 -> 42.9 | Helped |
| add pair self-attention to the mini-detector and split-cross-attention model (ablation) | C-DETR · COCO-val | AP: 42.9 -> 43.6 | Helped |
| allow decoder gradients to backpropagate to the mini-detector instead of stopping gradients (ablation) | DESTR · COCO-val | AP: 42.9 -> 42.1 | Hurt |
| remove the flip operator from pair self-attention (ablation) | DESTR · COCO-val | AP: 43.6 -> 42.5 | Hurt |
| use p instead of s in cross-attention (ablation) | DESTR · COCO-val | AP: 43.6 -> 41.8 | Hurt |
| increase testing queries from 100 to 200, with 300 training queries (ablation) | DESTR · COCO-val | AP: 41.8 -> 43.2 | Helped |
| increase testing queries from 200 to 300, with 300 training queries (ablation) | DESTR · COCO-val | AP: 43.2 -> 43.6 | Helped |
| increase testing queries from 300 to 400, with 300 training queries (ablation) | DESTR · COCO-val | AP: 43.6 -> 43.7 | Helped |
| increase training queries from 300 to 600, with 300 testing queries (ablation) | DESTR · COCO-val | AP: 43.6 -> 43.5 | No clear effect |
| increase queries from 300 to 600 in both training and testing (ablation) | DESTR · COCO-val | AP: 43.6 -> 43.8 | No clear effect |
Closest papers by meaning
- Anchor DETR: Query Design for Transformer-Based Detector (AAAI 2022 · similarity 0.89)
- DAC-DETR: Divide the Attention Layers and Conquer (NeurIPS 2023 · similarity 0.89)
- Sparse DETR: Efficient End-to-End Object Detection with Learnable Sparsity (ICLR 2022 · similarity 0.89)
- FP-DETR: Detection Transformer Advanced by Fully Pre-training (ICLR 2022 · similarity 0.88)
- Training Object Detectors From Scratch: An Empirical Study in the Era of Vision Transformer (CVPR 2022 · similarity 0.88)
- 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.