def test_step_end(self, outputs):
self.log("test_loss", outputs["loss"], on_step=True, on_epoch=True)
self.log(
"test_iou",
self.iou(outputs["preds"], outputs["target"]),
)
self.log(
"test_dice",
dice_score(outputs["preds"], outputs["target"]),
)
def validation_step(self, batch, batch_nb):
x, y = batch['image'], batch['mask']
y_hat = self.net(x)
loss = F.cross_entropy(y_hat, y) if self.n_classes > 1 else \
F.binary_cross_entropy_with_logits(y_hat, y)
dice_acc = dice_score(y_hat, y)
# logs metrics for each training_step,
# and the average across the epoch, to the progress bar and logger
self.log("val_acc", dice_acc, on_step=True, on_epoch=True, prog_bar=True, logger=True)
tensorboard_logs = {'train_loss': loss, "train_acc": dice_acc}
return {'val_loss': loss, 'log': tensorboard_logs}
def test_dice_score(pred, target, expected):
score = dice_score(tensor(pred), tensor(target))
assert score == expected