1、Improving Convolutional Networks with Self-calibrated Convolutions
论文地址：http://mftp.mmcheng.net/Papers/20cvprSCNet.pdf
代码地址：https://github.com/MCG-NKU/SCNet
2、DO-Conv: Depthwise Over-parameterized Convolutional Layer
地址：https://arxiv.org/pdf/2006.12030.pdf
github：https://github.com/yangyanli/DO-Conv.
3、Data-Driven Neuron Allocation for Scale Aggregation Networks
地址：https://arxiv.org/pdf/1904.09460.pdf
github：https://github.com/Eli-YiLi/ScaleNet
4、MixConv: Mixed Depthwise Convolutional Kernels
地址：https://arxiv.org/pdf/1907.09595.pdf
github：https://github.com/ tensorflow/tpu/tree/master/models/official/mnasnet/mixnet
5、 exploring self-attention for image recognition
地址：https://hszhao.github.io/papers/cvpr20_san.pdf
github：https://github.com/hszhao/SAN
6、MUXConv: Information Multiplexing in Convolutional Neural Networks
地址：https://arxiv.org/pdf/2003.13880.pdf
github：https://github.com/ human-analysis/MUXConv
7、Res2Net: A New Multi-scale Backbone Architecture
地址：https://arxiv.org/pdf/1904.01169.pdf
github：https://mmcheng.net/res2net
8、Rethinking Depthwise Separable Convolutions: How Intra-Kernel Correlations Lead to Improved MobileNets
地址：https://arxiv.org/pdf/2003.13549.pdf
9、DYNET: DYNAMIC CONVOLUTION FOR ACCELERATING CONVOLUTIONAL NEURAL NETWORKS
地址：https://arxiv.org/pdf/2004.10694.pdf
10、Dynamic Convolution: Attention over Convolution Kernels
地址：https://arxiv.org/pdf/1912.03458.pdf
11、CondConv: Conditionally Parameterized Convolutions for Efficient Inference
地址：https://arxiv.org/pdf/1904.04971.pdf
github：https://github.com/tensorflow/tpu/tree/master/ models/official/efficientnet/condconv
12、XSepConv: Extremely Separated Convolution
地址：https://arxiv.org/pdf/2002.12046.pdf
13、ACNet: Strengthening the Kernel Skeletons for Powerful CNN via Asymmetric Convolution Blocks
地址：https://arxiv.org/pdf/1908.03930.pdf
14、Drop an Octave: Reducing Spatial Redundancy in Convolutional Neural Networks with Octave Convolution
地址：https://arxiv.org/pdf/1904.05049.pdf
原文：https://www.cnblogs.com/xiximayou/p/13192378.html