The Feature Pyramid Network (FPN) presents a remarkable approach to alleviate the scale variance in object representation by performing instance-level assignments. Nevertheless, this strategy ignores the distinct characteristics of different sub-regions in an instance. To this end, we propose a fine-grained dynamic head to conditionally select a pixel-level combination of FPN features from different scales for each instance, which further releases the ability of multi-scale feature representation. Moreover, we design a spatial gate with the new activation function to reduce computational complexity dramatically through spatially sparse convolutions. Extensive experiments demonstrate the effectiveness and efficiency of the proposed method on several state-of-the-art detection benchmarks. Code is available at https://github.com/StevenGrove/DynamicHead.
@article{arxiv.2012.03519,
title = {Fine-Grained Dynamic Head for Object Detection},
author = {Lin Song and Yanwei Li and Zhengkai Jiang and Zeming Li and Hongbin Sun and Jian Sun and Nanning Zheng},
journal= {arXiv preprint arXiv:2012.03519},
year = {2020}
}