If you want to do the test of our model, it will cost about 10 minutes using GPU. Unfortunately, we are not able to support an online GPU. Also, we think it is not a good experience for you to wait for a long time on the website to get the result.
In addition, because we lost our trained model and it is time limited before we build this website so that we have not enough time to train this model.
So we publish our code on ++Github++. There are two kinds of code. Line-Artist-Code is the one without training. You can train this model by yourself using this code. Line-Artist-Code-Trained contains our trained model and you can use this code to do the test or as a pretrained model for your work.
Line-Artist-Code
It is a pity that we are not able to support an end-to-end model. So you need use the three models dividually.
SIE model
This model acts as the roll of generating training data for the later two models.
To do the smoothing, run smoothing.m in /Line-Artist-Code/SIE/
To generate sketch from a smoothed image, run sketching.m in /Line-Artist-Code/SIE/. Parameters’ meanings are shown in /Line-Artist-Code/SIE/PencilDrawing.m. You can change the pencil sample in /Line-Artist-Code/SIE/pencils/.
After running these two matlab code, you can get the sketches of training dataset.
DIS model
This model changes sketches into reality images. Just run train.py in /Line-Artist-Code/DIS/ to train this model and run test.py in the same directory to test if you have trained a suitable model.
AWAST model
Just run neural_style.py in /Line-Artist-Code/AWAST/ to generate your styled images.
Line-Artist-Code-Trained
We will publish the trained model in the near future.
近期评论