I'm working on a VAE model to produce synthetic data of X-Ray diffraction spectrums.

I try to figure out how I can measure the quality of the spectrums. The goal would be to produce synthetic data which is similar to the training data but also different from the training data. The spectrums should keep their characteristics, but should be different in terms of noise and intensity.

I trained models which can produce those type of spectrums (because I checked some of them visual), but I don't know how to quantify the difference/similarity to the origin (1) and the difference between the produced synthetic spectrums in one dataset (2).

Are there any methods to quantify these points?


Due to subjective nature, quantitative evaluation of synthetic images is difficult in general. However, there are metrics like Inception Score or FID score that are used for evaluation of generative models like GANs or VAEs. Technically, it considers two aspects of the generated data:

  1. Similarity with training data
  2. Diversity within itself

Even though such metrics do not assess new images as we humans do, but it is widely accepted in the community.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.