Abstract

Restoring high-quality images from blurred videos is a highly challenging task, especially in severely blurred scenes. In recent years, event-based methods have achieved significant progress in video deblurring. However, the modal differences between the event and image increase the difficulty of feature fusion. Additionally, the sparsity of event makes it difficult to restore some local details. To address these issues, we propose a new video deblurring method. Firstly, we design a cross modal collaborative attention mechanism to effectively fuse features from blurred frames and event frames, thereby deeply extracting motion information from event frames. Secondly, we utilize a diffusion model to generate spatial guiding prior feature, enhancing local details and textures. Furthermore, we propose an event-guided dynamic feature fusion module that adaptively integrates spatio-temporal information from neigh boring frames. Experimental results on both synthetic and real datasets demonstrate that our method outperforms the current state-of-the-art approaches. The code is available at: https://github.com/Frank-Zhou-01/EDVD-main

Fig. 1. Deblurring results of different methods on the GOPRO dataset. DSTNet rely solely on blurred frames to restore clear images and perform poorly in highly blurred scenes. EFNet can effectively restore highly blurred scenes with the help of event information. However, due to the sparsity of event information, it cannot effectively restore some texture details. Compared with previous methods, our method can restore clearer details and images.

Fig. 2. Overall structure of the deblurring network. It consists of three components: cross-modal collaborative attention (CMCA), dynamic spatial enhancement (DSE) module, and event-guided dynamic feature fusion (EGDFF) module. In processing a blurred video and its corresponding event sequence, CMCA mitigates severe motion blur by fusing image features and event features across multiple dimensions. Subsequently, guided by spatial prior generated from a diffusion model, DSE module further enhances details and textures. Finally, a bidirectional feature fusion mechanism is employed to fully utilize temporal information from the video sequence, thereby reconstructing clear frames.

https://doi.org/10.1109/TIP.2026.3684428