Image-to-Image - a neural network that turns doodles into "photos"
Image-to-Image - a neural network that turns doodles into "photos"
Anonim

American developer Christopher Hesse presented the Edges2cats project, which uses a neural network to turn sketches into pictures. To transform drawings, the program uses a database of several thousand photographs.

Image-to-Image - a neural network that turns doodles into "photos"
Image-to-Image - a neural network that turns doodles into "photos"

The user is prompted to draw a black and white picture with the mouse and press the Process button. The resulting image can be saved.

Image-to-Image: cats
Image-to-Image: cats

In addition to cats, on the site where the program is located, you can also draw a house, shoes or a bag - all using neural networks.

Image
Image
Image
Image
Image
Image

To implement his idea, Hesse used the Google TensorFlow library and the pix2pix program. The finished configuration with some changes is called pix2pix-tensorflow and is published in the repository on GitHub. A ready-made Docker container with already installed and configured programs is also available.

Recommended: