consulting

[daqiang2]

Hello martin!
The work is very great.But i want to consult whether FID can be used in medical image field? After all,Iception mode is trained on natural images.Thank you very much!

[mhex]

Hi, good question. From the theoretical point of view the only assumption for the FID is that the activations of the Inception coding layer (last 2048 dim pooling layer) are Gaussian distributed given your data. Then the FID calculates the squared Wasserstein-2 distance between the Gaussians of the activations of the real and generated data. From a practical point of view i suggest that also medical images produce near Gaussian activations, near enough such that both distributions, real and generated) can be compared.

[lzh1-hub]

Hello martin! The work is very great.But i want to consult whether FID can be used in medical image field? After all,Iception mode is trained on natural images.Thank you very much!

Hello, I am a student and I have the same problem as you. Have you solved the problem of calculating FID for medical images?