Clarification on FID and IS Values in Table 1

[haoweiz23]

Dear Authors,

I find the FID values in Table 1 a bit confusing. For example, D6 achieves the best FID but the worst IS value, while D1 shows the inverse trend. This is particularly perplexing because D6 also achieves a higher compression ratio.

Does this imply that a higher compression ratio tends to result in better generative performance due to the improved FID? I would appreciate your clarification on how to interpret these results.

Thank you for your time and insight!

[Probe100]

Thank you for pointing out this observation. We believe the discrepancy between IS and FID may be due to the inaccuracy of the FID metric, as indicated in previous research [1]. The FID does not fully capture the visual quality of the generated images. However, changes in the FID score can still be meaningful, as they reflect shifts in the distribution of the generated images.

[1] Jayasumana S, Ramalingam S, Veit A, et al. Rethinking fid: Towards a better evaluation metric for image generation[C]// CVPR. 2024: 9307-9315.

[DefTruth]

same question, the values of FID is very confuse.