Less is More: Focus Attention for Efficient DETR

TL;DR

Focus-DETR improves efficiency in DETR models by selectively attending to informative tokens using a dual attention encoder, achieving high accuracy with reduced computation.

Contribution

The paper introduces Focus-DETR, a novel approach that reconstructs the encoder with dual attention and token scoring to enhance efficiency without sacrificing detection performance.

Findings

Achieves 50.4 AP on COCO, outperforming similar sparse models.

Maintains comparable complexity to state-of-the-art sparse DETR variants.

Effectively reduces background queries and enhances object query semantics.

Abstract

DETR-like models have significantly boosted the performance of detectors and even outperformed classical convolutional models. However, all tokens are treated equally without discrimination brings a redundant computational burden in the traditional encoder structure. The recent sparsification strategies exploit a subset of informative tokens to reduce attention complexity maintaining performance through the sparse encoder. But these methods tend to rely on unreliable model statistics. Moreover, simply reducing the token population hinders the detection performance to a large extent, limiting the application of these sparse models. We propose Focus-DETR, which focuses attention on more informative tokens for a better trade-off between computation efficiency and model accuracy. Specifically, we reconstruct the encoder with dual attention, which includes a token scoring mechanism that considers both localization and category semantic information of the objects from multi-scale feature maps. We efficiently abandon the background queries and enhance the semantic interaction of the fine-grained object queries based on the scores. Compared with the state-of-the-art sparse DETR-like detectors under the same setting, our Focus-DETR gets comparable complexity while achieving 50.4AP (+2.2) on COCO. The code is available at https://github.com/huawei-noah/noah-research/tree/master/Focus-DETR and https://gitee.com/mindspore/models/tree/master/research/cv/Focus-DETR.