MWQ: Multiscale Wavelet Quantized Neural Networks

Abstract

Model quantization can reduce the model size and computational latency, it has become an essential technique for the deployment of deep neural networks on resourceconstrained hardware (e.g., mobile phones and embedded devices). The existing quantization methods mainly consider the numerical elements of the weights and activation values, ignoring the relationship between elements. The decline of representation ability and information loss usually lead to the performance degradation. Inspired by the characteristics of images in the frequency domain, we propose a novel multiscale wavelet quantization (MWQ) method. This method decomposes original data into multiscale frequency components by wavelet transform, and then quantizes the components of different scales, respectively. It exploits the multiscale frequency and spatial information to alleviate the information loss caused by quantization in the spatial domain. Because of the flexibility of MWQ, we demonstrate three applications (e.g., model compression, quantized network optimization, and information enhancement) on the ImageNet and COCO datasets. Experimental results show that our method has stronger representation ability and can play an effective role in quantized neural networks.

Wavelet Feature Maps Compression for Image-to-Image CNNs

“…Using a different transform (known or learned) is also possible, at the corresponding computational cost. [49] introduce quantization in the wavelet domain, as we do in this work. However, the authors suggest to improve the quantization scheme by learning a different clipping parameter per wavelet component, but without the feature shrinkage stage, which is the heart of our approach (we use the same clipping parameter for the whole layer using 8 bits for hardware efficiency).…”

“…However, the authors suggest to improve the quantization scheme by learning a different clipping parameter per wavelet component, but without the feature shrinkage stage, which is the heart of our approach (we use the same clipping parameter for the whole layer using 8 bits for hardware efficiency). As mentioned before, one can use our WCC together with different types of quantization schemes, in particular including the one proposed by [49], taking the additional hardware complexity of using different clipping parameter per component into account. [3,44,7].…”