NTIRE 2024 Challenge on Bracketing Image Restoration and Enhancement: Datasets, Methods and Results

Low-light photography presents significant challenges. Multi-image processing methods have made numerous attempts to obtain high-quality photos, yet remain unsatisfactory. Recently, bracketing image restoration and enhancement has received increased attention. By leveraging the full potential of multi-exposure images, several tasks (including denoising, deblurring, high dynamic range enhancement, and super-resolution) can be jointly addressed. This paper reviews the NTIRE 2024 challenge on bracketing image restoration and enhancement. In the challenge, participants are required to process multi-exposure RAW images to generate noise-free, blur-free, high dynamic range, and even higher-resolution RAW images. The challenge comprises two tracks. Track 1 does not incorporate the super-resolution task, whereas Track 2 does. Each track featured five teams participating in the final testing phase. The proposed methods establish new state-of-the-art performance benchmarks.

1. Introduction

Low-light photography is a widely desired yet challenging problem. Recent years have witnessed significant advancements in enhancing low-light image quality through learning-based methods. In comparison with single-image restoration (e.g., denoising [1, 9, 30, 45, 49, 94, 98, 99], deblurring [18], [62], [66], [74], [95], [97], and super-resolution (SR) [41, 46, 48, 51, 100, 102, 103]) and enhancement (e.g., high dynamic range (HDR) reconstruction [15], [26], [42], [55], [69], [110]), multi-image processing methods offer more advantages in mitigating the ill-posed nature of this problem and can generate results with higher fidelity.