Fréchet Inception Distance

The visual inspection of generated images is a great start. But given they look okay, a more precise, quantitative evaluation will be helpful to understand the generator's performance. You will evaluate your GAN using the Fréchet Inception Distance, or FID.

Two tensors with fake and real images, 32 examples each, are available to you as fake and real, respectively. Use them to compute the FID!

This exercise is part of the course

Deep Learning for Images with PyTorch

View Course

Exercise instructions

Import FrechetInceptionDistance from the appropriate torchmetrics module.
Instantiate the FID metric based on the 64th Inception feature layer and assign it to fid.
Update fid with real image tensor, multiplied by 255 and parsed to torch.uint8.
Compute the fid metric, assigning the output to fid_score.

Hands-on interactive exercise

Have a go at this exercise by completing this sample code.

# Import FrechetInceptionDistance
____

# Instantiate FID
fid = ____(____)

# Update FID with real images
fid.update((fake * 255).to(torch.uint8), real=False)
fid.update(____)

# Compute the metric
fid_score = ____
print(fid_score)

Edit and Run Code