Pytorch was lacking code to calculate the Inception Score for GANs. This repository fills this gap. However, we do not recommend using the Inception Score to evaluate generative models, see our note for why.
Clone the repository and navigate to it:
$ git clone git@github.com:sbarratt/inception-score-pytorch.git
$ cd inception-score-pytorch
To generate random 64x64 images and calculate the inception score, do the following:
$ python inception_score.py
The only function is inception_score. It takes a list of numpy images normalized to the range [0,1] and a set of arguments and then calculates the inception score. Please assure your images are 3x299x299 and if not (e.g. your GAN was trained on CIFAR), pass resize=True to the function to have it automatically resize using bilinear interpolation before passing the images to the inception network.
def inception_score(imgs, cuda=True, batch_size=32, resize=False, splits=1):
"""Computes the inception score of the generated images imgs
imgs -- Torch dataset of (3xHxW) numpy images normalized in the range [-1, 1]
cuda -- whether or not to run on GPU
batch_size -- batch size for feeding into Inception v3
splits -- number of splits
"""You will need torch, torchvision, numpy/scipy.
This project is licensed under the MIT License - see the LICENSE.md file for details
- Inception Score from Improved Techniques for Training GANs