Abstract
In this paper, we present an encoder-decoder architecture that exploits global and local semantics for the automatic image colorization problem. For the global semantics, the low-level encoding features are fine-tuned by the scene-context classification to integrate the global image style. Moreover, the architecture deals with the uncertainty and relations among the scene styles based on the label smoothing and pre-trained weights from Places365. For local semantics, three branches learn the mutual benefits at the pixel-level, in which average and multi-modal distributions are respectively created from regression and soft-encoding branches, while the segmentation branch determines to which object the pixel belongs. Our experiments, which involve training with the Coco-Stuff dataset and validation on DIV2K, Places365, and ImageNet, show that our results are very encouraging.
Highlights
Colorizing a gray-scale image brings a lot of special semantics into that image, and helps the image become more vivid and emotional [1]
The result shows that semantic segmentation played an important role in enhancing the colorization results, and it helped our method improve the accuracy of the ab channels
The image colorization at the pixel-level was fine-tuned by the semantic segmentation of both objects and the kinds of scenes from the Coco-Stuff dataset
Summary
Colorizing a gray-scale image brings a lot of special semantics into that image, and helps the image become more vivid and emotional [1]. The combination of global and local knowledge in scene semantic and semantic segmentation will help the observers determine what the objects are in the image and reference object models about the colors of the objects. The low-level encoding features will be fine-tuned by the scenecontext classification branch in the middle of the encoder and decoder branches This means they contain the encoding data of the input image, and the global style of the image, to help the decoder branch colorize images more precisely based on the global semantics of the scenecontext. Propose an encoder-decoder architecture that uses the scene-context classification and pixel-wise segmentation for image colorization.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
