question on fid50k_full metric after resuming training on a larger dataset

[MoemaMike]

I trained my model on a dataset of 2500 images for about 4 days and was getting good results with steady dropping fid50k_full metric. At that point i decided to enlarge the dataset to 7000 images. So i added 4500 to the original 2500 bringing total to 7000 and i resumed training with this superset.

Given the very heterogenous nature of this dataset i expected the fid50k_full metric to go way up initially upon the resumed training and then gradually decline again. But surprisingly the metric held steady and continued to decline very slowly .. which was a pleasant surprise but it makes me wonder if that is the expected behavior. i thought the additional images in the training set would present a much bigger challenge and cause it to spike up?