lllyasviel / sketchkeras Goto Github PK

I have been using this to convert a video into sketch animation but each image is resized into smaller image thus cannot retain quality. How can I retain image original size?

I tried this method and here are my results:

Original

After "sketched"

Can you give a hint on what's your training sets look like? Thank you.

There is no code in this repository that deals with creation of creating a neural net in tensorflow. Could you please upload it? Your results seem interesting.

不好意思,
我是剛從PaintsChainer 的issue 連接過來的, 你的項目很有趣, 可以加個好友嗎?

Hello, I want to use your pre-trained model in inference, but where can I access pretrained model (mod.h5)?

The result looks nice.
BUT, what's the dataset, how you trained the model?
Hope you could share some details

Can you briefly describe the dataset and training loss/algorithm used for the "sketchifying" neural network? I couldn't find any comments after reading this repo, paintschainer and your style2paints repo and paper. Many thanks.

I load the mod and print mod.InputLayers[0].input_shape and get the (None, 512, 512, 1).
But the code in main.py indicates that the input shape should be [None, 512, 512, 3] or [3, 512, 512, None] depending on the data format.
When I feed [3, 512, 512, batch_size] or [batch_size, 512, 512, 3] to mod.predict, Exception is raised.
Can you help me ? What's going on here?

I can't find the keras model mod.h5

Hi, your u-net architecture works very well. Would you mind sharing the training data(or how to collect this data) and your u-net architecture?

Thank you!