基于TensorFlow 2.x的一些CNN模块/网络的实现

开源一些基于TensorFlow 2.x的CNN模块/网络的实现，可能不定时更新。仓库链接：TensorFlow-2-Implementations-of-CNN-Based-Networks

目前的实现包括：

Feature Extraction/Fusion Blocks

• Atrous Convolutional Block for 1D (data points / sequences) or 2D inputs (images / feature maps), suggested by An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling

• Receptive Field Block, from Receptive Field Block Net for Accurate and Fast Object Detection

Attention Blocks

• Squeeze-and-Excitation Block (Kind of Channel Attention), from Squeeze-and-Excitation Networks

• Convolutional Block Attention Module (CBAM), including Channel Attention Module and Spatial Attention Module, from CBAM: Convolutional Block Attention Module

• Non-Local Block, including ‘Gaussian’, ‘Embedded Gaussian’, ‘Dot Product’ and ‘Concatenation’ modes, from Non-local Neural Networks

• Dual Attention Module, including Channel Attention Module and Position Attention Module, from Dual Attention Network for Scene Segmentation

Backbone Networks

• DenseNet, from Densely Connected Convolutional Networks

参考了以下文章/仓库中的一些代码实现，在此感谢：

[1] https://github.com/philipperemy/keras-tcn
[2] https://github.com/Baichenjia/Tensorflow-TCN/blob/master/tcn.py
[3] https://arxiv.org/pdf/1803.01271.pdf
[4] https://arxiv.org/pdf/1711.07767.pdf
[5] https://arxiv.org/abs/1709.01507
[6] https://github.com/kobiso/CBAM-tensorflow-slim/blob/master/nets/attention_module.py
[7] https://arxiv.org/abs/1807.06521
[8] https://arxiv.org/pdf/1711.07971.pdf
[9] https://github.com/titu1994/keras-non-local-nets/blob/master/non_local.py
[10] https://github.com/Tramac/Non-local-tensorflow/tree/master/non_local
[11] https://arxiv.org/pdf/1809.02983.pdf
[12] https://github.com/niecongchong/DANet-keras/blob/master/layers/attention.py
[13] https://github.com/okason97/DenseNet-Tensorflow2/blob/master/densenet/densenet.py
[14] https://arxiv.org/pdf/1608.06993.pdf