Recently, learning-based multi-exposure fusion (MEF) methods have made significant improvements. However, these methods mainly focus on static scenes and are prone to generate ghosting artifacts when tackling a more common scenario, i.e., the input images include motion, due to the lack of a benchmark dataset and solution for dynamic scenes. In this paper, we fill this gap by creating an MEF dataset of dynamic scenes, which contains multi-exposure image sequences and their corresponding high-quality reference images. To construct such a dataset, we propose a 'static-for-dynamic' strategy to obtain multi-exposure sequences with motions and their corresponding reference images. To the best of our knowledge, this is the first MEF dataset of dynamic scenes. Correspondingly, we propose a deep dynamic MEF (DDMEF) framework to reconstruct a ghost-free high-quality image from only two differently exposed images of a dynamic scene. DDMEF is achieved through two steps: pre-enhancement-based alignment and privilege-information-guided fusion. The former pre-enhances the input images before alignment, which helps to address the misalignments caused by the significant exposure difference. The latter introduces a privilege distillation scheme with an information attention transfer loss, which effectively improves the deghosting ability of the fusion network. Extensive qualitative and quantitative experimental results show that the proposed method outperforms state-of-the-art dynamic MEF methods. The source code and dataset are released at https://github.com/Tx000/Deep_dynamicMEF.

Challenging motion, which tends to cause artifacts, is a key problem in the video denoising task. Recent video denoising methods have attempted to address this problem. However, they usually provide general performance evaluation on the overall dataset and cannot provide a comprehensive analysis for the influence of different motion levels. Thus, we questioned whether these methods can effectively deal with different scene motions. To this end, we synthesize a dataset containing videos with different motion levels and capture a new dataset that consists of videos involving large-scale motion. Then, we provide a comprehensive analysis on the elaborately collected datasets and find that, as the motion level increases, the performance of the denoising models based on implicit motion estimation (IME) declines sharply, while explicit motion estimation (EME) contributes to a more robust denoising quality. Therefore, in this work, we present an EME-embedded progressive denoising framework that fully considers the relationship between the noise removal and motion estimation. Specifically, we decouple video denoising into spatial denoising, EME-based frame reconstruction, and temporal refining processes. Spatial denoising improves the accuracy of EME process in the case of videos suffering from heavy noise, while the temporal refining process refines the denoised frame by utilizing temporal redundancy of the reconstructed motion-free frames. Extensive experiments demonstrate that the proposed method outperforms existing state-of-the-art methods, especially for videos containing large-scale motion.