AIMIND proposed as a drop-in replacement for FID in generative-model evaluation
The paper’s authors say MIND (Monge Inception Distance) requires 10 times fewer samples than FID, is more robust and is faster to compute.
TLDR
The authors introducing MIND argue that FID requires too many samples, slowing evaluations of generative models. They present MIND—Monge Inception Distance—as a drop-in replacement, claiming it needs 10 times fewer samples while offering greater robustness and faster computation.
MIND proposed as a drop-in replacement for FID in generative-model evaluation
The paper’s authors say MIND (Monge Inception Distance) requires 10 times fewer samples than FID, is more robust and is faster to compute.
TLDR
The authors introducing MIND argue that FID requires too many samples, slowing evaluations of generative models. They present MIND—Monge Inception Distance—as a drop-in replacement, claiming it needs 10 times fewer samples while offering greater robustness and faster computation.
