Learned Smartphone ISP on Mobile GPUs with Deep Learning, Mobile AI & AIM 2022 Challenge: Report

TL;DR

This paper reports on the development of efficient deep learning-based image signal processing pipelines for mobile GPUs, capable of real-time high-fidelity image enhancement on smartphones, based on the AIM 2022 challenge.

Contribution

It introduces novel end-to-end AI models optimized for mobile GPU deployment, demonstrating real-time processing of high-quality images using TensorFlow Lite.

Findings

Models process Full HD images in under 50 ms

Achieved high fidelity image quality comparable to professional cameras

Solutions are compatible with recent mobile GPUs

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.