Inconsistent Training loss in python_window and detail_view

[Felix10546]

Hi, 

I used the python command to execute GP "Train Deep Learning Model".

Here is my display in python window(left) and the detail_view(right) of the geoprocessing tool:

Clearer image is in the attachment.

PS.

I am doing a mask rcnn training with 2 classes.

Do anyone know which one is true?

The validation loss is consistent but the training loss isn't.

I want to attach training graph in my result with correct values.

Thank a lot if anyone can help with the answer.

[Samaloysius]

Hi, 

For each training run , the learning rate as well as real time augmentation is different hence the training/validations losses will be different for different runs. As long as the validation loss is decreasing with every epoch for a specific run the training is successful.