kvsnoufal/gauganpytorch

Paper To Code implementation of NVIDIA's GauGan on a custom Landscape 's Dataset. Generating photorealistic-ish:p images from drawings

GauGan: Generating Photorealistic Images from Drawings

Paper To Code implementation of NVIDIA's GauGan on a custom Landscape 's Dataset. Generating photorealistic-ish:p images from drawings

Origina Paper Semantic Image Synthesis with Spatially-Adaptive Normalization

Gaugan uses a special normalization technique for improving the quality of the data. The generator is capable of taking as input a semantic map (a drawing) and generating a photorealistic image as the output. Further it is also capable of multimodal image synthesis - which means, it can generate images in various different styles. So for the same drawing, it can generate multiple images.

In my implementation, I downloaded a dataset of landscape images from kaggle and used a pretrained semantic segmentation model (deeplab v2) to generate semantic maps of the image. This is how I compiled the dataset.

Results:

Shoulders of Giants:

1. Semantic Image Synthesis with Spatially-Adaptive Normalization (https://arxiv.org/abs/1903.07291)
2. Official Github Implementation : https://github.com/NVlabs/SPADE
3. Implementation in Keras : https://keras.io/examples/generative/gaugan/
4. Flickr Landscape Dataset: https://www.kaggle.com/datasets/arnaud58/landscape-pictures
5. DeepLab model for semantic segmentation: https://github.com/kazuto1011/deeplab-pytorch