Research map / Detection and segmentation / CNN features
Mr. DETR++: Instructive Multi-Route Training for Detection Transformers with Mixture-of-Experts
Chang-Bin Zhang, Yujie Zhong, Kai Han
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 auxiliary route with independent FFN (ablation) | Deformable-DETR++ · COCO 2017 val ablation (R50, 12 ep, 900 queries) | AP: 47.6 -> 49.6 | Improved |
| add auxiliary route with instructive self-attention (ablation) | Deformable-DETR++ · COCO 2017 val ablation (R50, 12 ep, 900 queries) | AP: 47.6 -> 50.4 | Improved |
| combine both auxiliary routes (Mr. DETR) (ablation) | Deformable-DETR++ · COCO 2017 val ablation (R50, 12 ep, 900 queries) | AP: 47.6 -> 50.7 | Improved |
| replace FFN route with route-aware MoE (ablation) | Mr. DETR · COCO 2017 val ablation (R50, 12 ep, 900 queries) | AP: 50.7 -> 51.0 | Improved |
| add scale-aware MoE in encoder (ablation) | Mr. DETR++ without calibration · COCO 2017 val ablation (R50, 12 ep, 900 queries) | AP: 51.0 -> 51.4 | Improved |
| add localization-aware score calibration (ablation) | Mr. DETR++ · COCO 2017 val ablation (R50, 12 ep, 900 queries) | AP: 51.4 -> 51.8 | Improved |
| instruct by concatenation instead of independent self-attention in Route-3 (ablation) | Mr. DETR · COCO 2017 val ablation (R50, 12 ep, 900 queries) | AP: 50.0 -> 50.7 | Improved |
| instruct by concatenation instead of removing self-attention (ablation) | Mr. DETR · COCO 2017 val ablation (R50, 12 ep, 900 queries) | AP: 49.9 -> 50.7 | Improved |
| instruct by concatenation instead of addition (ablation) | Mr. DETR · COCO 2017 val ablation (R50, 12 ep, 900 queries) | AP: 50.1 -> 50.7 | Improved |
| apply MoE on x64 scale features in encoder (ablation) | Mr. DETR++ · COCO 2017 val ablation (R50, 12 ep, 900 queries) | AP: 51.3 -> 51.8 | Improved |
| extend encoder MoE to x32 and x16 scales (ablation) | Mr. DETR++ · COCO 2017 val ablation (R50, 12 ep, 900 queries) | AP: 51.8 -> 51.9 | No clear effect |
| apply MoE to all four scales (ablation) | Mr. DETR++ · COCO 2017 val ablation (R50, 12 ep, 900 queries) | AP: 51.9 -> 51.5 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- Deformable-DETR++ — Deformable DETR: Deformable Transformers for End-to-End Object Detection (ICLR 2021) · better on 4 of 4 shared results
- MS-DETR — MS-DETR: Efficient DETR Training with Mixed Supervision (CVPR 2024) · better on 3 of 3 shared results
- Relation-DETR — Relation DETR: Exploring Explicit Position Relation Prior for Object Detection (ECCV 2024) · 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
- Mr. DETR: Instructive Multi-Route Training for Detection Transformers (CVPR 2025 · similarity 0.96)
- Group DETR: Fast DETR Training with Group-Wise One-to-Many Assignment (ICCV 2023 · similarity 0.90)
- End-to-End Object Detection with Transformers (ECCV 2020 · similarity 0.90)
- MS-DETR: Efficient DETR Training with Mixed Supervision (CVPR 2024 · similarity 0.88)
- DECO: Unleashing the Potential of ConvNets for Query-based Detection and Segmentation (ICLR 2025 · similarity 0.87)
- DetKDS: Knowledge Distillation Search for Object Detectors (ICML 2024 · 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.
