AdaConv

Unofficial PyTorch implementation of the Adaptive Convolution architecture for image style transfer from "Adaptive Convolutions for Structure-Aware Style Transfer". Disclaimer: I have not trained the model the full number of iterations yet, this is still a work in progress.

Direct link to the adaconv module.

Direct link to the kernel predictor module.

Data

Trained with MS COCO train2017 dataset for content images and WikiArt train dataset for style images. By default the content images should be placed in ./data/MSCOCO/train2017 and the style images in ./data/WikiArt/train. You can change these directories by passing arguments when running the script: python .\train.py --content= --style= or by changing the default value in /lib/lightning/datamodule.py

Architecture (from the original paper):

Preliminary results after training 45k iterations: