generated image classification model

I´ve used ML.NET model generation via that user interface and it generated me model with 0,93% microaccuracy which is really great.. but when I tried to retrain the model (with the code it generated)
this was its accuracy output it had average microaccuracy 0.12% or something like that, even training took 10 less minutes. Am I missing something obvious? or am I misunderstanding the output data on that picture?