Neural-Style-Transfer

PyTorch implementation of A Neural Algorithm of Artistic Style by Leon A. Gatys, et al.

Progress

This is the progress of the first few iterations

progress

progress

Implementation Details

The hyperparameters are same as used in the paper.
All the code was written and ran on Google Colab.
As in the paper, conv1_1, conv2_1, conv3_1, conv4_1, conv5_1 are used for style loss.
SImilarly, conv4_2 is used for content loss.
PyTorch’s forward hooks are used to extract intermediate feature maps.
VGG19’s MaxPool2d layers were replaced by AvgPool2d layers.

Requirements

torch
torchvision
numpy
matplotlib
Pillow
imageio

References

Leon A. Gatys, et al. A Neural Algorithm of Artistic Style [arxiv]