Conditional DETR for Fast Training Convergence

TL;DR

This paper introduces conditional DETR, a modification to the original DETR model that significantly accelerates training convergence by using a conditional spatial query mechanism, reducing reliance on high-quality content embeddings.

Contribution

The paper proposes a novel conditional cross-attention mechanism in DETR that speeds up training convergence by narrowing the spatial focus of attention heads.

Findings

Conditional DETR converges 6.7x faster with R50 and R101 backbones.

It achieves 10x faster convergence with stronger backbones DC5-R50 and DC5-R101.

The method maintains competitive detection performance while significantly reducing training time.

Abstract

The recently-developed DETR approach applies the transformer encoder and decoder architecture to object detection and achieves promising performance. In this paper, we handle the critical issue, slow training convergence, and present a conditional cross-attention mechanism for fast DETR training. Our approach is motivated by that the cross-attention in DETR relies highly on the content embeddings for localizing the four extremities and predicting the box, which increases the need for high-quality content embeddings and thus the training difficulty. Our approach, named conditional DETR, learns a conditional spatial query from the decoder embedding for decoder multi-head cross-attention. The benefit is that through the conditional spatial query, each cross-attention head is able to attend to a band containing a distinct region, e.g., one object extremity or a region inside the object box. This narrows down the spatial range for localizing the distinct regions for object classification and box regression, thus relaxing the dependence on the content embeddings and easing the training. Empirical results show that conditional DETR converges 6.7x faster for the backbones R50 and R101 and 10x faster for stronger backbones DC5-R50 and DC5-R101. Code is available at https://github.com/Atten4Vis/ConditionalDETR.