Skip to content

Add Fréchet Codec Distance Metric #71

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 7 commits into
base: magpietts_2503
Choose a base branch
from

Conversation

rfejgin
Copy link

@rfejgin rfejgin commented Apr 30, 2025

No description provided.

@rfejgin
Copy link
Author

rfejgin commented Apr 30, 2025

This PR introduces the Fréchet Codec Distance metric. This is a new metric we are experimenting with which is based on the Fréchet Inception Distance (FID).

We compute the distance between the distribution of real codec frames and generated codec frames. The distance is computed in the (dequantized) embedding space of the codec.

Note that FD distances are computed at the dataset level. The number of real and generated frames should large for a reliable metric.

The metric is now called from infer_and_evaluate.py and included in our set of reported metrics.

@rfejgin rfejgin marked this pull request as ready for review May 1, 2025 04:32
@rfejgin rfejgin changed the title Frechet Codec Distance metric Add Frechet Codec Distance Metric May 1, 2025
@rfejgin rfejgin changed the title Add Frechet Codec Distance Metric Add Fréchet Codec Distance Metric May 1, 2025
rfejgin added 3 commits April 30, 2025 22:20
Turns out there is an API for this for both FSQ and RVQ quantizers -- so we use it here.
@rfejgin
Copy link
Author

rfejgin commented May 7, 2025

I have resolved conflicts with latest magpietts_2503. This is ready for review / merge.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant