Published on:
8 April 2024
Primary Category:
Computer Vision and Pattern Recognition
Paper Authors:
Xiaoyan Cong,
Yue Wu,
Qifeng Chen,
Chenyang Lei
Proposes automatic colorization framework enabling iterative editing of results
Key module is Imagination Module generating multiple reference images
Mimics human creativity and imagination in coloring process
Allows localized changes to results by modifying reference images
Achieves state-of-the-art flexibility and editability
Imagination-based automatic image colorization
The authors propose a framework for automatic image colorization that allows iterative editing by generating multiple colorized images as references to guide the colorization. A key module is the Imagination Module, which leverages diffusion models to synthesize reference images containing diverse and realistic colors based on understanding the content and structure of the input grayscale image. This mimics human imagination and creativity in the colorization process. A Reference Refinement Module then selects the optimal reference composition to colorize the input. Compared to previous automatic colorization methods, this framework allows localized edits and modifications of the results by changing the reference images.
No comments yet, be the first to start the conversation...
Sign up to comment on this paper