Fréchet Inception Distance

Tensorflow implementation of the "Fréchet Inception Distance" (FID) between two image distributions, along with a numpy interface. The FID can be used to evaluate generative models by calculating the FID between real and fake data distributions (lower is better).

Major Dependencies

• tensorflow==1.14 or (tensorflow==1.15 and tensorflow-gan==1.0.0.dev0) or (tensorflow>=2 and tensorflow-gan>=2.0.0)

Features

• Fast, easy-to-use and memory-efficient
• No prior knowledge about Tensorflow is necessary if your are using CPUs or GPUs
• Makes use of TF-GAN
• Downloads InceptionV1 automatically
• Compatible with both Python 2 and Python 3

Usage

• If you are working with GPUs, use fid.py; if you are working with TPUs, use fid_tpu.py and pass a Tensorflow Session and a TPUStrategy as additional arguments.
• Call get_fid(images1, images2), where images1, images2 are numpy arrays with values ranging from 0 to 255 and shape in the form [N, 3, HEIGHT, WIDTH] where N, HEIGHT and WIDTH can be arbitrary. dtype of the images is recommended to be np.uint8 to save CPU memory.
• A smaller BATCH_SIZE reduces GPU/TPU memory usage, but at the cost of a slight slowdown.
• If you want to compute a general "Fréchet Classifier Distance" with activations (e.g., outputs of the last pooling layer) act1 and act2 from another classifier, call activations2distance(act1, act2). act1 and act2 can be numpy arrays of a same arbitrary shape [N, d].

Examples

GPU:

TPU and TF1:

TPU and TF2:

Links

• The Fréchet Inception Distance was proposed in the paper GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium
• Code for the Inception Score