Rethinking FID: Towards a Better Evaluation Metric for Image Generation
This addresses the problem of unreliable evaluation metrics for image generation models, which is crucial for researchers and practitioners in computer vision and generative AI, though it is incremental as it builds on existing metric frameworks.
The paper highlights significant drawbacks of the Frechet Inception Distance (FID) as an evaluation metric for image generation, showing it contradicts human raters and fails to capture model improvements, and proposes CMMD as a more robust alternative based on CLIP embeddings and maximum mean discrepancy.
As with many machine learning problems, the progress of image generation methods hinges on good evaluation metrics. One of the most popular is the Frechet Inception Distance (FID). FID estimates the distance between a distribution of Inception-v3 features of real images, and those of images generated by the algorithm. We highlight important drawbacks of FID: Inception's poor representation of the rich and varied content generated by modern text-to-image models, incorrect normality assumptions, and poor sample complexity. We call for a reevaluation of FID's use as the primary quality metric for generated images. We empirically demonstrate that FID contradicts human raters, it does not reflect gradual improvement of iterative text-to-image models, it does not capture distortion levels, and that it produces inconsistent results when varying the sample size. We also propose an alternative new metric, CMMD, based on richer CLIP embeddings and the maximum mean discrepancy distance with the Gaussian RBF kernel. It is an unbiased estimator that does not make any assumptions on the probability distribution of the embeddings and is sample efficient. Through extensive experiments and analysis, we demonstrate that FID-based evaluations of text-to-image models may be unreliable, and that CMMD offers a more robust and reliable assessment of image quality.