LineCloser
LineCloser copied to clipboard
Unofficial Keras implementation for paper 'Joint Gap Detection and Inpainting of Line Drawings'.
LineCloser
Unofficial Keras implementation of Joint Gap Detection and Inpainting of Line Drawings.
Overview
Joint gap for line-drawings. Model1 uses network from the paper. For stable training, BN was added for all Conv2D. Model2 uses common network for inpaint.
Dependencies
- Keras2 (Tensorflow backend)
- OpenCV3
- CairoSVG
Usage
-
Set up directories.
-
Download the model from release and put it in the same folder with code.
-
Run
predict.py
for prediction. Runmodel{NUM}.py
for train.
Data Preparation
There are 3 methods for data generation, DATA_GEN
, DATA_GAP
and DATA_THIN
.
-
Use
DATA_GEN
for training, the data is generated online. -
Collect line-drawings with LineDistiller.
-
Put line-drawings into
data/line
, usingDATA_GAP
for training. -
Thin(normalize) the line-drawings with LineNormalizer or tranditional thinning method.
-
Manually processe line-drawings and thinning results(threshold etc.), then crop them into pieces.
-
Put line-drawings into
data/line
and put thinning results intodata/thin
, usingDATA_THIN
for training.
Models
Models are licensed under a CC-BY-NC-SA 4.0 international license.
From Project HAT by Hepesu With :heart: