search

mit-quest / necstlab-damage-segmentation

MIT License
5 stars 6 forks source link

argument to control trained model saving #102

Closed CarolinaFurtado closed 3 years ago

CarolinaFurtado commented 3 years ago

Added a --metric-checkpoint argument, default to 'loss' that controls based on what variable, a trained model is saved

if 'loss' is selected: image

if val_loss is selected:

image

if nothing is selected, default to loss: image