LineFiller

Leak-proofing line art filling and segmentation based on trapped-ball filling.

The Full pipeline including different deep learning models and algorithm. For now, the repository only including trapped-ball filling(segmentation).

1. Input is be prepocessed by deep learning model LineCloser and LineThinner, these result a nice normalization line-drawing. Work in progess
2. A combination of Trapped-ball and FloodFill algorithm are used to do segmentation, the raw output will be processed by simple merge operation. Done
3. For each segment, a deep learning model LineFeature is used to extract features, then a cluster algorithm is used for grouping. Work in progess
4. Using LineFeature for classification or comparing to determine the color palatte by exmpale or user input. Work in progess

Extra pipelien

1. Using model from JACS for color palatte design. Work in progess
2. Using LineShader for shading. Work in progess
3. Using a model for coloring stylization. Work in progess

Example

The line art is drawn with reference from Violet Evergarden. <img src="example.png"/>

Result using trapped-ball and floodfill pipeline. Only floodfill leads result with leak. <img src="fills.png"/>

Result after merged. <img src="fills_merged.png"/>

Result after performed line thinning. <img src="fills_merged_no_contour.png"/>

From Project HAT by Hepesu With :heart: