An Efficient and Rotation Invariant Fourier-Based Metric for Assessing the Quality of Images Created by Generative Models

Recent progress in generative image modeling is leading to a new era of high-resolution fakes visually indistinguishable from real life images. However, the development of metrics capable of discerning whether images are synthetic or not runs behind the race of achieving the best generator, thus bringing potential threats. We propose a rotation invariant metric capable of distinguishing real and generated image datasets and we call it CSD (Circular Spectrum Distance) due to its circular nature and its inherent relation to the Fourier Spectrum. Its performance is analysed on a whole brain MRI dataset. CSD has similar behavior to FID during training but requires smaller batch sizes and is faster to compute.