Non-official + minimal reimplementation of HoloGAN by Nguyen-Phuoc, et al: https://arxiv.org/abs/1904.01326
This repo is a loose reimplementation of HoloGAN, originally by Nguyen-Phuoc et al: https://arxiv.org/abs/1904.01326
I do not claim or guarantee any correctness of this implementation. This was implemented indepedently without consulting any of the original authors of the paper or other code.
First, download the CelebA dataset, extract the images inside img_align_celeba
to some directory, and export the environment variable DATASET_CELEBA
to point to this folder (for instance, by running the command export DATASET_CELEBA=/datasets/celeba/img_align_celeba
).
Then, run python task_launcher.py
. To run the example training script, cd into exps
and run example.sh
.
Here is an example set of interpolations at 200 epochs.
Download the Cars dataset. You will need to put this dataset in a folder with the environment variable DATASET_CARS
pointing to it, and using --dataset=cars
in the task launcher. For instance:
cd some_directory
wget http://imagenet.stanford.edu/internal/car196/cars_train.tgz
wget http://imagenet.stanford.edu/internal/car196/cars_test.tgz
tar -xvzf cars_train.tgz
tar -xvzf cars_test.tgz
Here I bullet point some caveats and things that are perhaps worth noting if you are running HoloGAN experiments.