Papers › Mr. DETR: Instructive Multi-Route Training for Detection Transformers

Mr. DETR: Instructive Multi-Route Training for Detection Transformers

13 Dec 2024CVPR 2025 1arXiv:2412.10028archive 2025-07-28

Chang-Bin Zhang, Yujie Zhong, Kai Han

Existing methods enhance the training of detection transformers by incorporating an auxiliary one-to-many assignment. In this work, we treat the model as a multi-task framework, simultaneously performing one-to-one and one-to-many predictions. We investigate the roles of each component in the transformer decoder across these two training targets, including self-attention, cross-attention, and feed-forward network. Our empirical results demonstrate that any independent component in the decoder can effectively learn both targets simultaneously, even when other components are shared. This finding leads us to propose a multi-route training mechanism, featuring a primary route for one-to-one prediction and two auxiliary training routes for one-to-many prediction. We enhance the training mechanism with a novel instructive self-attention that dynamically and flexibly guides object queries for one-to-many prediction. The auxiliary routes are removed during inference, ensuring no impact on model architecture or inference cost. We conduct extensive experiments on various baselines, achieving consistent improvements as shown in Figure 1. Project page: https://visual-ai.github.io/mrdetr

PaperPDFConference PDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

Visual-AI/Mr.DETR mentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

DecoderObject DetectionPrediction

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Object Detection COCO 2017 val Mr. DETR (Swin-L, 1x, 5cale) AP 61.8 #1 of 33 Archive leaderboard report
Object Detection COCO 2017 val Mr. DETR (Swin-L, 1x, 5cale) AP50 79.0 #1 of 33 Archive leaderboard report
Object Detection COCO 2017 val Mr. DETR (Swin-L, 1x, 5cale) AP75 67.6 #1 of 33 Archive leaderboard report
Object Detection COCO 2017 val Mr. DETR (Swin-L, 1x, 5cale) APL 75.7 #1 of 33 Archive leaderboard report
Object Detection COCO 2017 val Mr. DETR (Swin-L, 1x, 5cale) APM 65.6 #1 of 33 Archive leaderboard report
Object Detection COCO 2017 val Mr. DETR (Swin-L, 1x, 5cale) APS 47.7 #1 of 33 Archive leaderboard report
Object Detection COCO 2017 val Mr. DETR (Swin-L, 1x, 4scale) AP 58.4 #2 of 33 Archive leaderboard report
Object Detection COCO 2017 val Mr. DETR (Swin-L, 1x, 4scale) AP50 76.3 #2 of 33 Archive leaderboard report
Object Detection COCO 2017 val Mr. DETR (Swin-L, 1x, 4scale) AP75 63.9 #2 of 33 Archive leaderboard report
Object Detection COCO 2017 val Mr. DETR (Swin-L, 1x, 4scale) APL 75.3 #2 of 33 Archive leaderboard report
Object Detection COCO 2017 val Mr. DETR (Swin-L, 1x, 4scale) APM 62.8 #2 of 33 Archive leaderboard report
Object Detection COCO 2017 val Mr. DETR (Swin-L, 1x, 4scale) APS 40.8 #2 of 33 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections