Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DetNAS: Backbone Search for Object Detection

Yukang Chen, Tong Yang, Xiangyu Zhang, Gaofeng Meng, Xinyu Xiao, Jian Sun

TL;DR

DetNAS tackles the problem of suboptimal backbone design for object detection by introducing a dedicated backbone search framework using one‑shot NAS and an evolutionary search strategy. It decouples weight training from architecture optimization via a three‑step pipeline: pretraining a one‑shot supernet on ImageNet, fine‑tuning on detection data, and EA‑based backbone search, with careful handling of BN statistics. The resulting DetNASNet backbones achieve higher mmAP than hand‑crafted networks at similar or lower FLOPs across COCO and VOC, and exhibit detection‑specific architectural patterns distinct from networks optimized for image classification. The approach demonstrates that task‑specific backbone search is both feasible and beneficial for real‑world detectors, with implications for more efficient and effective object detection systems.

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.

DetNAS: Backbone Search for Object Detection

TL;DR

DetNAS tackles the problem of suboptimal backbone design for object detection by introducing a dedicated backbone search framework using one‑shot NAS and an evolutionary search strategy. It decouples weight training from architecture optimization via a three‑step pipeline: pretraining a one‑shot supernet on ImageNet, fine‑tuning on detection data, and EA‑based backbone search, with careful handling of BN statistics. The resulting DetNASNet backbones achieve higher mmAP than hand‑crafted networks at similar or lower FLOPs across COCO and VOC, and exhibit detection‑specific architectural patterns distinct from networks optimized for image classification. The approach demonstrates that task‑specific backbone search is both feasible and beneficial for real‑world detectors, with implications for more efficient and effective object detection systems.

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.

Paper Structure

This paper contains 15 sections, 4 equations, 5 figures, 8 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Object Detection
  4. Neural Architecture Search
  5. Detection Backbone Search
  6. Motivation
  7. Our NAS Pipeline
  8. Search Space Design
  9. Search Algorithm
  10. Experimental Settings
  11. Experimental Results
  12. Main Results
  13. Ablation Studies
  14. DetNAS Architecture and Discussions
  15. Conclusion

Figures (5)

  • Figure 1: The pipeline of DetNAS that searches for backbones in object detectors. There are three steps: supernet pre-training on ImageNet, supernet fine-tuning on the detection training set, e.g., COCO, and architecture search on the trained supernet with the evolution algorithm. The validation set is actually split from COCO trainval35k and consists of 5k images.
  • Figure 2: Curve of EA and Random during search.
  • Figure 3: Random models on COCO-FPN.
  • Figure 4: The searched architecture pattern comparison in the small (20 blocks) search space. From top to bottom, they are ClsNASNet, DetNAS (COCO-FPN) and DetNAS (COCO-RetinaNet).
  • Figure 5: DetNASNet architecture