Anchor-Based Plain Net for Mobile Image Super-Resolution

Zongcai Du, Jie Liu, Jie Tang, Gangshan Wu; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2021, pp. 2494-2502

Along with the rapid development of real-world applications, higher requirements on the accuracy and efficiency of image super-resolution (SR) are brought forward. Though existing methods have achieved remarkable success, the majority of them demand high computational resources and large amount of RAM, and thus they can not be well applied to mobile device. In this paper, we aim at designing efficient architecture for 8-bit quantization and deploy it on mobile device. First, we conduct an experiment about meta-node latency by decomposing lightweight SR architectures, which determines the portable operations we can utilize. Then, we dig deeper into what kind of architecture is beneficial to 8-bit quantization and propose anchor-based plain net (ABPN). Finally, we adopt quantization-aware training to further boost the performance. Our INT8 quantization model can even achieve nearly the same performance as the floating-point network, with only 0.07dB drop.