CBNetV2: A Composite Backbone Network Architecture for Object Detection

Modern top-performing object detectors depend heavily on backbone networks, whose advances bring consistent performance gains through exploring more effective network structures. In this paper, we propose a novel and flexible backbone framework, namely CBNetV2, to better train existing open-sourced pre-trained backbones under the pre-training fine-tuning protocol. In particular, CBNetV2 architecture groups multiple identical backbones, which are connected through composite connections. Specifically, it integrates the high- and low-level features of multiple backbone networks and gradually expands the receptive field to more efficiently perform object detection. We also propose a better training strategy with assistant supervision for CBNet-based detectors. CBNetV2 has strong generalization capabilities for different backbones and head designs of the detector architecture. Without additional pre-training, CBNetV2 can be adapted to various backbones, including manual-based and NAS-based, as well as CNN-based and Transformer-based ones. Experiments provide strong evidence showing that composite backbones are more efficient, effective, and resource-friendly than wider and deeper networks. CBNetV2 is compatible with the head designs of most mainstream detectors, including one-stage and two-stage detectors, as well as anchor-based and anchor-free-based ones, and significantly improve their performances by more than 3.0% AP over the baseline on COCO. Particularly, under the single-model and single-scale testing protocol, our Dual-Swin-L achieves 59.4% box AP and 51.6% mask AP on COCO test-dev, which is significantly better than the state-of-the-art result (i.e., 57.7% box AP and 50.2% mask AP). Code is available at https://github.com/VDIGPKU/CBNetV2.