[CVPR2024]DCNv4 Efficient Deformable ConvNets_efficient deformable convnets: rethinking dynamic

Basic Information:

Title: Efficient Deformable ConvNets: Rethinking Dynamic and Sparse Operator for Vision Applications (高效的可变形卷积神经网络：重新思考动态和稀疏算子在视觉应用中的作用)
Authors: Yuwen Xiong, Zhiqi Li, Yuntao Chen, Feng Wang, Xizhou Zhu, Jiapeng Luo, Wenhai Wang, Tong Lu, Hongsheng Li, Yu Qiao, Lewei Lu, Jie Zhou, Jifeng Dai
Affiliation: University of Toronto, OpenGVLab, Shanghai AI Laboratory, Nanjing University, CAIR, HKISI, CAS, Tsinghua University, SenseTime Research, The Chinese University of Hong Kong
Keywords: Deformable Convolution, Vision Applications, Dynamic Property, Expressive Power, Memory Optimization, Speed Increase

论文简要

本文提出了一种高效且有效的算子，名为Deformable Convolution v4 (DCNv4)，用于广泛的视觉应用。通过两个关键改进，DCNv4解决了其前身DCNv3的局限性，包括**增强动态性