Deep Colorization: A Channel Attention-based CNN for Video Colorization
Ye Yang, Yao Liu, Hui Yuan, Yanhan Chu · 2022
We investigate the video colorization problem which aims to convert a grayscale video to a colorful version. Video colorization is a difficult problem due to the temporal and spatial correlation between video frames. It can be widely used in many fields, such as restoring black-and-white film back to color and cross color prediction for video coding. We propose an end-to-end framework to solve the video colorization task. Our work is based on temporal convolutional neural networks with attention mechanisms. The proposed method can control the contribution of each channel when extracting features. Our network can colorize multiple frames at the same time and can handle long video sequences colorization by giving only one frame as a reference. Quantitative analysis shows that our method outperforms existing approaches on the tested video sequences.