Optimize Deep Convolutional Neural Network with Ternarized Weights and High Accuracy
IEEE Winter Conference on Applications of Computer Vision (WACV) 2019
https://github.com/elliothe/Ternarized_Neural_Network Pytorch
本文只对权重进行三值量化,每个卷积层和全连接层可以使用多个三值网络层表示以便提升三值网络的性能
III. W EIGHT T ERNARIZATION T RAINING M ETHOD
A. Ternarization with iterative statistical scaling
权重的三值量化:
其中 scaling factor 为
量化阈值为
dot-product operations 计算如下