在移动NPU，移动AI和AIM 2022挑战上有效，准确的量化图像超分辨率：报告

论文标题

在移动NPU，移动AI和AIM 2022挑战上有效，准确的量化图像超分辨率：报告

Efficient and Accurate Quantized Image Super-Resolution on Mobile NPUs, Mobile AI & AIM 2022 challenge: Report

论文作者

Ignatov, Andrey, Timofte, Radu, Denna, Maurizio, Younes, Abdel, Gankhuyag, Ganzorig, Huh, Jingang, Kim, Myeong Kyun, Yoon, Kihwan, Moon, Hyeon-Cheol, Lee, Seungho, Choe, Yoonsik, Jeong, Jinwoo, Kim, Sungjei, Smyl, Maciej, Latkowski, Tomasz, Kubik, Pawel, Sokolski, Michal, Ma, Yujie, Chao, Jiahao, Zhou, Zhou, Gao, Hongfan, Yang, Zhengfeng, Zeng, Zhenbing, Zhuge, Zhengyang, Li, Chenghua, Zhu, Dan, Sun, Mengdi, Duan, Ran, Gao, Yan, Kong, Lingshun, Sun, Long, Li, Xiang, Zhang, Xingdong, Zhang, Jiawei, Wu, Yaqi, Pan, Jinshan, Yu, Gaocheng, Zhang, Jin, Zhang, Feng, Ma, Zhe, Wang, Hongbin, Cho, Hojin, Kim, Steve, Li, Huaen, Ma, Yanbo, Luo, Ziwei, Li, Youwei, Yu, Lei, Wen, Zhihong, Wu, Qi, Fan, Haoqiang, Liu, Shuaicheng, Zhang, Lize, Zong, Zhikai, Kwon, Jeremy, Zhang, Junxi, Li, Mengyuan, Fu, Nianxiang, Ding, Guanchen, Zhu, Han, Chen, Zhenzhong, Li, Gen, Zhang, Yuanfan, Sun, Lei, Zhang, Dafeng, Yang, Neo, Liu, Fitz, Zhao, Jerry, Ayazoglu, Mustafa, Bilecen, Bahri Batuhan, Hirose, Shota, Arunruangsirilert, Kasidis, Ao, Luo, Leung, Ho Chun, Wei, Andrew, Liu, Jie, Liu, Qiang, Yu, Dahai, Li, Ao, Luo, Lei, Zhu, Ce, Hong, Seongmin, Park, Dongwon, Lee, Joonhee, Lee, Byeong Hyun, Lee, Seunggyu, Chun, Se Young, He, Ruiyuan, Jiang, Xuhao, Ruan, Haihang, Zhang, Xinjian, Liu, Jing, Gendy, Garas, Sabor, Nabil, Hou, Jingchao, He, Guanghui

论文摘要

图像超分辨率是移动设备和IoT设备上的常见任务，在该任务中，通常需要高档并增强低分辨率图像和视频帧。尽管过去已经为此问题提出了许多解决方案，但它们通常与具有许多计算和内存约束的低功率移动NPU不兼容。在此移动AI挑战中，我们解决了此问题，并建议参与者设计有效的量化图像超分辨率解决方案，该解决方案可以在移动NPU上展示实时性能。向参与者提供了DIV2K数据集和经过训练的INT8模型，以进行高质量的3X图像升级。在突触VS680智能主板上评估了所有模型的运行时，具有专用的Edge NPU，能够加速量化的神经网络。所有提出的解决方案都与上述NPU完全兼容，在重建全高清分辨率图像时，表明高达60 fps的速率。本文提供了挑战中所有模型的详细描述。

Image super-resolution is a common task on mobile and IoT devices, where one often needs to upscale and enhance low-resolution images and video frames. While numerous solutions have been proposed for this problem in the past, they are usually not compatible with low-power mobile NPUs having many computational and memory constraints. In this Mobile AI challenge, we address this problem and propose the participants to design an efficient quantized image super-resolution solution that can demonstrate a real-time performance on mobile NPUs. The participants were provided with the DIV2K dataset and trained INT8 models to do a high-quality 3X image upscaling. The runtime of all models was evaluated on the Synaptics VS680 Smart Home board with a dedicated edge NPU capable of accelerating quantized neural networks. All proposed solutions are fully compatible with the above NPU, demonstrating an up to 60 FPS rate when reconstructing Full HD resolution images. A detailed description of all models developed in the challenge is provided in this paper.

下载PDF全文

下载文献需遵守相关版权规定

论文标题