Mr. DETR++: Instructive Multi-Route Training for Detection Transformers with Mixture-of-Experts

TL;DR

Mr. DETR++ introduces a multi-route training framework with a novel instructive self-attention and Mixture-of-Experts to enhance detection transformers, achieving consistent improvements across multiple segmentation tasks.

Contribution

It proposes a multi-route training mechanism with instructive self-attention and MoE, enabling effective multi-target learning and knowledge sharing in detection transformers.

Findings

Consistent performance improvements across detection, instance, and panoptic segmentation.

Effective multi-task learning with shared components in the transformer decoder.

Flexible framework adaptable to various vision tasks.

Abstract

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 propose a novel instructive self-attention mechanism, integrated into the first auxiliary route, which dynamically and flexibly guides object queries for one-to-many prediction. For the second auxiliary route, we introduce a route-aware Mixture-of-Experts (MoE) to facilitate knowledge sharing while mitigating potential conflicts between routes. Additionally, we apply an MoE to low-scale features in the encoder, optimizing the balance between efficiency and effectiveness. The auxiliary routes are discarded during inference. We conduct extensive experiments across various object detection baselines, achieving consistent improvements as demonstrated in Fig. 1. Our method is highly flexible and can be readily adapted to other tasks. To demonstrate its versatility, we conduct experiments on both instance segmentation and panoptic segmentation, further validating its effectiveness. Project page: https://visual-ai.github.io/mrdetr/