Learned Smartphone ISP on Mobile GPUs with Deep Learning, Mobile AI & AIM 2022 Challenge: Report
Abstract
The role of mobile cameras increased dramatically over the past few years, leading to more and more research in automatic image quality enhancement and RAW photo processing. In this Mobile AI challenge, the target was to develop an efficient end-to-end AI-based image signal processing (ISP) pipeline replacing the standard mobile ISPs that can run on modern smartphone GPUs using TensorFlow Lite. The participants were provided with a large-scale Fujifilm UltraISP dataset consisting of thousands of paired photos captured with a normal mobile camera sensor and a professional 102MP medium-format FujiFilm GFX100 camera. The runtime of the resulting models was evaluated on the Snapdragon's 8 Gen 1 GPU that provides excellent acceleration results for the majority of common deep learning ops. The proposed solutions are compatible with all recent mobile GPUs, being able to process Full HD photos in less than 20-50 milliseconds while achieving high fidelity results. A detailed description of all models developed in this challenge is provided in this paper.
Cite
@article{arxiv.2211.03885,
title = {Learned Smartphone ISP on Mobile GPUs with Deep Learning, Mobile AI & AIM 2022 Challenge: Report},
author = {Andrey Ignatov and Radu Timofte and Shuai Liu and Chaoyu Feng and Furui Bai and Xiaotao Wang and Lei Lei and Ziyao Yi and Yan Xiang and Zibin Liu and Shaoqing Li and Keming Shi and Dehui Kong and Ke Xu and Minsu Kwon and Yaqi Wu and Jiesi Zheng and Zhihao Fan and Xun Wu and Feng Zhang and Albert No and Minhyeok Cho and Zewen Chen and Xiaze Zhang and Ran Li and Juan Wang and Zhiming Wang and Marcos V. Conde and Ui-Jin Choi and Georgy Perevozchikov and Egor Ershov and Zheng Hui and Mengchuan Dong and Xin Lou and Wei Zhou and Cong Pang and Haina Qin and Mingxuan Cai},
journal= {arXiv preprint arXiv:2211.03885},
year = {2022}
}