Recently, many deep learning applications have been used on the mobile platform. To deploy them in the mobile platform, the networks should be quantized. The quantization of computer vision networks has been studied well but there have been few studies for the quantization of image restoration networks. In this paper, we studied the effect of the quantization of activations for deep learning network on image quality following previous study for weight quantization for deep learning network. This study is also about the quantization on raw RGBW image demosaicing for 10 bit image while fixing weight bit as 8 bit. Experimental results show that 11 bit activation quantization can sustain image quality at the similar level with floating-point network. Even though the activations bit-depth can be very small bit in the computer vision applications, but image restoration tasks like demosaicing require much more bits than those applications. 11 bit may not fit the general purpose hardware like NPU, GPU or CPU but for the custom hardware it is very important to reduce its hardware area and power as well as memory size.
Latest trend in image sensor technology allowing submicron pixel size for high-end mobile devices comes at very high image resolutions and with irregularly sampled Quad Bayer color filter array (CFA). Sustaining image quality becomes a challenge for the image signal processor (ISP), namely for demosaicing. Inspired by the success of deep learning approach to standard Bayer demosaicing, we aim to investigate how artifacts-prone Quad Bayer array can benefit from it. We found that deeper networks are capable to improve image quality and reduce artifacts; however, deeper networks can be hardly deployed on mobile devices given very high image resolutions: 24MP, 36MP, 48MP. In this article, we propose an efficient end-to-end solution to bridge this gap—a duplex pyramid network (DPN). Deep hierarchical structure, residual learning, and linear feature map depth growth allow very large receptive field, yielding better details restoration and artifacts reduction, while staying computationally efficient. Experiments show that the proposed network outperforms state of the art for standard and Quad Bayer demosaicing. For the challenging Quad Bayer CFA, the proposed method reduces visual artifacts better than state-of-the-art deep networks including artifacts existing in conventional commercial solutions. While superior in image quality, it is 2–25 times faster than state-of-the-art deep neural networks and therefore feasible for deployment on mobile devices, paving the way for a new era of on-device deep ISPs.