Hi,
congratulations on your great work. I was wondering whether the scripts for computing FID and sFID are correct. Since the paper does not mention the number of images used in computing these metrics on ImageNet (Table 4.), I assume that you used all examples of two specific classes, e.g. Zebra and Sorrel, to compute FID, since this would give a bit over 2k images, the minimum required for classical FID. However, for sFID, this number is too small to allow for computing the metric, as each split contains only around 1k images. Could you provide more details about how these metrics were computed and which images were used for that? Thank you in advance.
Hi,
congratulations on your great work. I was wondering whether the scripts for computing FID and sFID are correct. Since the paper does not mention the number of images used in computing these metrics on ImageNet (Table 4.), I assume that you used all examples of two specific classes, e.g. Zebra and Sorrel, to compute FID, since this would give a bit over 2k images, the minimum required for classical FID. However, for sFID, this number is too small to allow for computing the metric, as each split contains only around 1k images. Could you provide more details about how these metrics were computed and which images were used for that? Thank you in advance.