DAQ: Channel-Wise Distribution-Aware Quantization for Deep Image Super-Resolution Networks
This work provides a method to accurately quantize deep image super-resolution networks to ultra-low precisions without extensive fine-tuning, which is significant for deploying these models on resource-constrained devices.
This paper addresses the problem of quantizing deep convolutional neural networks for image super-resolution, particularly at ultra-low precisions (4 bits or lower), where existing methods suffer significant performance drops or require extensive fine-tuning. The proposed Distribution-Aware Quantization (DAQ) scheme enables accurate training-free quantization, outperforming recent training-free and even training-based methods on state-of-the-art image super-resolution networks in ultra-low precision.
Quantizing deep convolutional neural networks for image super-resolution substantially reduces their computational costs. However, existing works either suffer from a severe performance drop in ultra-low precision of 4 or lower bit-widths, or require a heavy fine-tuning process to recover the performance. To our knowledge, this vulnerability to low precisions relies on two statistical observations of feature map values. First, distribution of feature map values varies significantly per channel and per input image. Second, feature maps have outliers that can dominate the quantization error. Based on these observations, we propose a novel distribution-aware quantization scheme (DAQ) which facilitates accurate training-free quantization in ultra-low precision. A simple function of DAQ determines dynamic range of feature maps and weights with low computational burden. Furthermore, our method enables mixed-precision quantization by calculating the relative sensitivity of each channel, without any training process involved. Nonetheless, quantization-aware training is also applicable for auxiliary performance gain. Our new method outperforms recent training-free and even training-based quantization methods to the state-of-the-art image super-resolution networks in ultra-low precision.