Research map / Detection and segmentation / Open-set and few-shot detection
DisCo DETR: Distance-aware Multi-view Contrastive Learning for DETR Pre-training
Chao Ouyang, Yuyang Bai, Jun Zhang, Tianlu Gao, Jun Hao, Lijun Kong, David Wenzhong Gao
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 distance-aware multi-view object query fusion (DMOQF) to Siamese DETR (ablation) | Siamese DETR · PASCAL VOC, Conditional DETR | AP: 50.3 -> 51.4 | Improved |
| add contrastive learning on decoder outputs (CLD) (ablation) | Siamese DETR · PASCAL VOC, Conditional DETR | AP: 50.3 -> 51.14 | Improved |
| combine DMOQF and CLD (ablation) | DMOQF only · PASCAL VOC, Conditional DETR | AP: 51.4 -> 51.7 | Improved |
| combine DMOQF and CLD (ablation) | DMOQF only · PASCAL VOC, Deformable-DETR MS | AP: 49.8 -> 50.2 | Improved |
| use projection and prediction heads in CLD (vs none) (ablation) | DisCo DETR · PASCAL VOC, Conditional DETR 100 queries | AP: 43.9 -> 47.8 | Improved |
| add prediction head fpred on top of fproj (ablation) | CLD · PASCAL VOC, Conditional DETR 100 queries | AP: 47.5 -> 47.8 | Improved |
| pre-train with DisCo DETR instead of Siamese DETR | DAB-DETR · COCO, 20/24 schedule | AP: 30.5 -> 33.9 | Improved |
| pre-train with DisCo DETR instead of DETReg | RT-DETR · COCO, 20/24 schedule | AP: 40.5 -> 46.7 | Improved |
| pre-train with DETReg instead of training from scratch | RT-DETR · PASCAL VOC | AP: 59.8 -> 59.0 | Hurt |
| pre-train with DisCo DETR instead of from scratch | RT-DETR · PASCAL VOC | AP: 59.8 -> 61.6 | Improved |
| pre-train with DisCo DETR instead of from scratch | Deformable-DETR single-scale · PASCAL VOC | AP: 38.5 -> 36.6 | Hurt |
| pre-train with DisCo DETR instead of from scratch | RT-DETR · COCO, 40/50 schedule | AP: 49.2 -> 49.8 | 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.
- DETReg — DETReg: Unsupervised Pretraining With Region Priors for Object Detection (CVPR 2022) · better on 2 of 2 shared results
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
- Decoupled DETR: Spatially Disentangling Localization and Classification for Improved End-to-End Object Detection (ICCV 2023 · similarity 0.89)
- DetCo: Unsupervised Contrastive Learning for Object Detection (ICCV 2021 · similarity 0.89)
- Semantic-Aligned Matching for Enhanced DETR Convergence and Multi-Scale Feature Fusion (IJCV 2024 · similarity 0.87)
- Siamese DETR (CVPR 2023 · similarity 0.87)
- Conditional DETR for Fast Training Convergence (ICCV 2021 · similarity 0.87)
- MS-DETR: Efficient DETR Training with Mixed Supervision (CVPR 2024 · 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.
