How to monitor a validation loss or metric? about pytorch-segmentation HOT 3 CLOSED

you can monitor any metric in the returned dict log in _valid_epoch. For instance to monitor the validation loss, try replacing "monitor": "max Mean_IoU" with "monitor": "min val_loss"

from pytorch-segmentation.

I think it is useful to change the validation metrics name to add some prefix like: "val_" . Maybe adding self.wrt_mode before the metric name in _get_seg_metrics funtion. Example: self.wrt_mode + "_Pixel_Accuracy". Because if we set the checkpoint monitor for "max Pixel_Accuracy" it will save the best model also using the training metric.

from pytorch-segmentation.

Only the validation metrics are tracked, the checking is executed only after an eval was conducted.

from pytorch-segmentation.