I am working on a unsupervised image reconstruction task. so i need to verify whether the reconstructed image by the model is good are not. I cant do that visually.
so i was calculating the sum of mean and standard deviation of all the predictions of the dataset and finding a threshold.
So that prediction loss greater than the threshold is not good and less than threshold is good for a given reconstructed image.
threshold = np.mean(losses_train_all)+np.std(losses_train_all)
I was trying to improve this threshold. are their any existence methods that suit my problem?