DetNAS: Backbone Search for Object Detection

TL;DR

This paper introduces DetNAS, a neural architecture search framework tailored for designing backbone networks optimized for object detection, outperforming traditional classification-based backbones on COCO with fewer FLOPs.

Contribution

DetNAS presents a novel NAS framework that efficiently searches for object detection-specific backbones using a supernet trained with detection tasks, bridging the gap between classification and detection.

Findings

Searched architectures outperform classification-based backbones on COCO.

Detection-specific backbones achieve higher accuracy with less FLOPs.

Framework is effective for both one-stage and two-stage detectors.

Abstract

Object detectors are usually equipped with backbone networks designed for image classification. It might be sub-optimal because of the gap between the tasks of image classification and object detection. In this work, we present DetNAS to use Neural Architecture Search (NAS) for the design of better backbones for object detection. It is non-trivial because detection training typically needs ImageNet pre-training while NAS systems require accuracies on the target detection task as supervisory signals. Based on the technique of one-shot supernet, which contains all possible networks in the search space, we propose a framework for backbone search on object detection. We train the supernet under the typical detector training schedule: ImageNet pre-training and detection fine-tuning. Then, the architecture search is performed on the trained supernet, using the detection task as the guidance. This framework makes NAS on backbones very efficient. In experiments, we show the effectiveness of DetNAS on various detectors, for instance, one-stage RetinaNet and the two-stage FPN. We empirically find that networks searched on object detection shows consistent superiority compared to those searched on ImageNet classification. The resulting architecture achieves superior performance than hand-crafted networks on COCO with much less FLOPs complexity.