shitagaki
lab-see-through
AIComputer VisionDiffusion ModelsImage SegmentationAnimeGenerative AI
View on GitHub 28
+340
// summary
See-through is a framework capable of automatically converting static anime illustrations into actionable 2.5D models. This method uses deep learning technology to decompose a single image into up to 23 semantically clear and in-painted layers, while inferring a reasonable drawing order. The project aims to provide Live2D artists with efficient auxiliary tools, simplifying the tedious process of manual segmentation and occlusion in-painting.
// use cases
01
Automatically decompose a single anime character image into 23 semantic layers including hair, face, clothing, and other parts.
02
Export PSD files containing complete layer information to facilitate subsequent editing and processing.
03
Utilize depth information and semantic segmentation technology to provide an efficient pre-processing workflow for anime character production.