A Pragmatic Note on Evaluating Generative Models with Fr\'echet Inception Distance for Retinal Image Synthesis

TL;DR

This paper critically examines the use of FID, a popular generative model evaluation metric, in retinal imaging, highlighting its limitations and advocating for task-specific assessments in biomedical image synthesis.

Contribution

The paper reveals the limitations of FID in biomedical retinal image synthesis and emphasizes the importance of downstream task evaluation for more meaningful assessment.

Findings

FID can misalign with task-specific performance in retinal image synthesis.

Metrics like FID may not reliably evaluate synthetic biomedical images for downstream tasks.

Task-based evaluation provides more relevant insights than FID alone.

Abstract

Fr\'echet Inception Distance (FID), computed with an ImageNet pretrained Inception-v3 network, is widely used as a state-of-the-art evaluation metric for generative models. It assumes that feature vectors from Inception-v3 follow a multivariate Gaussian distribution and calculates the 2-Wasserstein distance based on their means and covariances. While FID effectively measures how closely synthetic data match real data in many image synthesis tasks, the primary goal in biomedical generative models is often to enrich training datasets ideally with corresponding annotations. For this purpose, the gold standard for evaluating generative models is to incorporate synthetic data into downstream task training, such as classification and segmentation, to pragmatically assess its performance. In this paper, we examine cases from retinal imaging modalities, including color fundus photography and optical coherence tomography, where FID and its related metrics misalign with task-specific evaluation goals in classification and segmentation. We highlight the limitations of using various metrics, represented by FID and its variants, as evaluation criteria for these applications and address their potential caveats in broader biomedical imaging modalities and downstream tasks.