Core Concepts
Exposure bracketing unifies image restoration and enhancement tasks by leveraging multi-exposure images.
Abstract
The content discusses the challenges of capturing high-quality photos in low-light environments and proposes exposure bracketing as a solution. It introduces a novel approach that utilizes multi-exposure images for denoising, deblurring, HDR imaging, and super-resolution tasks. The method involves pre-training on synthetic data and self-supervised adaptation to real-world images. Extensive experiments show superior performance compared to existing methods.
Structure:
Introduction to the challenge of capturing clear photos in low-light conditions.
Comparison of various multi-image processing methods.
Proposal of exposure bracketing for unifying restoration and enhancement tasks.
Description of the proposed method using TMRNet and self-supervised adaptation.
Data simulation pipeline for synthesizing pairs and collecting real-world images.
Experimental results showcasing the method's performance against state-of-the-art approaches.
Stats
Burst Denoising: 37.01/0.9454/0.127
Burst SR: 28.59/0.8516/0.292