Research map / Vision-language and multimodal / Cross-modal learning
Semantic-Aligned Matching for Enhanced DETR Convergence and Multi-Scale Feature Fusion
Gongjie Zhang, Zhipeng Luo, Jiaxing Huang, Shijian Lu, Eric P. Xing
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 Semantics Aligner with max-pool resampling to DETR (ablation) | DETR · COCO val2017, 12 epochs, R50 (ablation) | AP: 22.3 -> 27.0 | Improved |
| use max-pool instead of average-pool query resampling (ablation) | SAM-DETR · COCO val2017, 12 epochs, R50 (ablation) | AP: 25.2 -> 27.0 | Improved |
| search one representative keypoint instead of max-pool (ablation) | SAM-DETR · COCO val2017, 12 epochs, R50 (ablation) | AP: 27.0 -> 28.6 | Improved |
| search 8 keypoints instead of 1 (ablation) | SAM-DETR · COCO val2017, 12 epochs, R50 (ablation) | AP: 28.6 -> 32.0 | Improved |
| add feature reweighting with previous queries (ablation) | SAM-DETR · COCO val2017, 12 epochs, R50 (ablation) | AP: 32.0 -> 33.1 | Improved |
| restrict keypoint search to reference boxes instead of whole image (ablation) | SAM-DETR · COCO val2017, 12 epochs, R50 (ablation) | AP: 30.0 -> 33.1 | Improved |
| remove dropout in Transformer (ablation) | SAM-DETR · COCO val2017, 12 epochs, R50 (ablation) | AP: 33.1 -> 34.2 | Improved |
| add multi-scale feature fusion with aligner (ablation) | SAM-DETR · COCO val2017, 12 epochs, R50 (ablation) | AP: 34.2 -> 41.9 | Improved |
| multi-scale fusion without semantic alignment (ablation) | DETR · COCO val2017, 12 epochs, R50 (ablation) | AP: 41.9 -> 29.1 | Hurt |
| add SMCA to SAM-DETR++ | SAM-DETR++ · COCO val2017, 12 epochs, multi-scale R50 | AP: 41.9 -> 43.2 | Improved |
| add DN training to SAM-DETR++ w/ SMCA | SAM-DETR++ · COCO val2017, 12 epochs, multi-scale R50 | AP: 43.2 -> 44.8 | Improved |
| add SAM-DETR to DETR under 12 epochs R50 | DETR · COCO val2017, 12 epochs, single-scale R50 | AP: 22.3 -> 34.2 | Improved |
Closest papers by meaning
- Accelerating DETR Convergence via Semantic-Aligned Matching (CVPR 2022 · similarity 0.98)
- Semi-DETR: Semi-Supervised Object Detection With Detection Transformers (CVPR 2023 · similarity 0.91)
- FeatAug-DETR: Enriching One-to-Many Matching for DETRs With Feature Augmentation (TPAMI 2024 · similarity 0.90)
- Semantic-Aligned Fusion Transformer for One-Shot Object Detection (CVPR 2022 · similarity 0.90)
- Mean Teacher DETR with Masked Feature Alignment: A Robust Domain Adaptive Detection Transformer Framework (AAAI 2024 · similarity 0.88)
- Semantic-Aware Transformation-Invariant RoI Align (AAAI 2024 · similarity 0.88)
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.
