metrics calculation in the https://colab.research.google.com/github/CompVis/taming-transformers/blob/master/scripts/reconstruction_usage.ipynb#scrollTo=ZQKgzqrywDKa

[kritiagg]

trafficstars

I was trying to calculate the PSNR, SSIM, FID, LPIPS metrics for the Dalle model using 256x256 imagenet validation set but I am getting very different numbers.

Can you please share more details on the metrics calculation:

• code for the 4 metrics (PSNR, SSIM, LPIPS, R-FID)
• What was the backend model used for LPIPS (alexnet or vggnet)
• What is R-FID? How is it different than FID?
• Also, from the metrics shown in the collab notebook, it looks like only R-FID metric is showing that the VQGAN model is better. Although, the dalle model images are definitely much worse.