This code is used for latent space visualization in Image-to-Image problems. Firstly, all images are passed through ResNet101 to extract features, which are then reduced dimension by t-SNE algorithm for plotting
To visualize latent space of n domain. All images of each domain are put into a folder.python --path [path] --label [label] --path_save result_path --n_sample number_of_sample --n_components number_of_component --perplexity perplexity --n_iter
where --path is a list of path to domain, --label is a list name of domain
python --path path/to/Monet_folder path/to/Photo_folder --label Monet_style Photo --n_sample 500 --n_components 2 --perplexity 40 --n_iter 4000
