Research map / Detection and segmentation / CNN features
FeatAug-DETR: Enriching One-to-Many Matching for DETRs With Feature Augmentation
Rongyao Fang, Peng Gao, Aojun Zhou, Yingjie Cai, Si Liu, Jifeng Dai, Hongsheng Li
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 FeatAug-FC | DAB-DETR-DC5 (R50) · COCO val2017 | AP: 45.7 -> 47.1 | Improved |
| add FeatAug-FC | Deform-DETR w/ tricks (R50, 12 epochs) · COCO val2017 | AP: 47.0 -> 48.7 | Improved |
| add FeatAug-FC | Deform-DETR w/ tricks (Swin-L, 12 epochs) · COCO val2017 | AP: 54.5 -> 55.8 | Improved |
| add DataAug (N=2, 48 epochs) | Deform-DETR w/ tricks (R50) · COCO val2017 | AP: 49.0 -> 50.0 | Improved |
| add FeatAug-Flip | H-Deformable-DETR (Swin-L, 900 queries) · COCO val2017 | AP: 57.9 -> 58.3 | Improved |
| use spatial flip/crop instead of image resizing only in DataAug (ablation) | Deform-DETR w/ tricks (R50) · COCO val2017 | AP: 48.9 -> 49.8 | Improved |
| increase DataAug augmentation count from N=2 to N=3 (ablation) | Deform-DETR w/ tricks (R50, 36 epochs) · COCO val2017 | AP: 49.8 -> 49.4 | Hurt |
| add FeatAug-FC on top of FeatAug-Flip (ablation) | Deform-DETR w/ tricks (R50, 24 epochs) · COCO val2017 | AP: 49.6 -> 49.9 | Improved |
| augment backbone features instead of encoder features (ablation) | FeatAug-Flip on Deform-DETR w/ tricks (R50, 12 epochs) · COCO val2017 | AP: 46.5 -> 48.4 | Improved |
| use separate projector for cropped features (ablation) | FeatAug-Crop on Deform-DETR w/ tricks (R50, 24 epochs) · COCO val2017 | AP: 49.2 -> 49.6 | Improved |
| add FeatAug-FC | Deformable-DETR one-stage (R50, 50 epochs) · COCO val2017 | AP: 44.5 -> 46.3 | Improved |
| use FeatAug-Flip instead of FeatAug-FC with Hybrid Matching (ablation) | H-Deformable-DETR (R50, 24 epochs) · COCO val2017 | AP: 50.1 -> 50.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Deformable-DETR — Deformable DETR: Deformable Transformers for End-to-End Object Detection (ICLR 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DETRs With Hybrid Matching (CVPR 2023 · similarity 0.92)
- Semantic-Aligned Matching for Enhanced DETR Convergence and Multi-Scale Feature Fusion (IJCV 2024 · similarity 0.90)
- Accelerating DETR Convergence via Semantic-Aligned Matching (CVPR 2022 · similarity 0.90)
- MS-DETR: Efficient DETR Training with Mixed Supervision (CVPR 2024 · similarity 0.90)
- CP-DETR: Concept Prompt Guide DETR Toward Stronger Universal Object Detection (AAAI 2025 · similarity 0.87)
- MI-DETR: An Object Detection Model with Multi-time Inquiries Mechanism (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.
