Saw this while training something. Since the graph automatically expands to include the highest point... well, if you get one image in a set that's a particular outlier, especially early on in the training process, that's going to make it rather hard to read a trend much lower down.
Saw this while training something. Since the graph automatically expands to include the highest point... well, if you get one image in a set that's a particular outlier, especially early on in the training process, that's going to make it rather hard to read a trend much lower down.