Awesome
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
andtensorflow-gan==1.0.0.dev0
) or (tensorflow>=2
andtensorflow-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, usefid_tpu.py
and pass a Tensorflow Session and a TPUStrategy as additional arguments. - Call
get_fid(images1, images2)
, whereimages1
,images2
are numpy arrays with values ranging from 0 to 255 and shape in the form[N, 3, HEIGHT, WIDTH]
whereN
,HEIGHT
andWIDTH
can be arbitrary.dtype
of the images is recommended to benp.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
andact2
from another classifier, callactivations2distance(act1, act2)
.act1
andact2
can be numpy arrays of a same arbitrary shape[N, d]
.
Examples
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