metadata

title: StyleTransfer
emoji: 🐢
colorFrom: indigo
colorTo: red
sdk: gradio
sdk_version: 4.27.0
app_file: app.py
pinned: false

Neural Style Transfer Deployment

DALI 2024S - Machine Learning Track

Deployment of the neural style transfer algorithm. Users can input their own style and content images and train the algorithm with their own hyperparameters. If you need a pair of images to expiriment with, I reccomend President.jpg as the content image and SelfPortrait.jpg as the style image. I get good results with 5 epochs, a learning rate of 0.01, a content loss of 1, and a style loss of 0.01.
Importantly, this algorithm is running on a CPU so keep training time in mind when setting the number of epochs.

Files

app.py: A copy of the interface I defined in the Python notebook
vgg19-dcbb9e9d.pth: The vgg19 pre-trained weights (so I don't have to query them each time I load the model)
President.jpg: Sample content image
SelfPortrait.jpg: Sample style image