Tensorboard

During the training, you can open TensorBoard and it will show the following statistics and visualization:

tensorboard --logdir=./log_sc

TensorBoard - Validation Metrics

These are the metrics computed for the validation dataset. It contains results for pixel-wise accuracy, Dice score, mIoU (mean intersection over union), pixel-wise precision, recall and specificity.

TensorBoard - Training set samples

These are visualizations of the training samples (after data augmentation), their ground truths and predictions from the network.

TensorBoard - Validation set samples

These are visualizations of the validation samples, their ground truths and predictions from the network.

TensorBoard - Losses

This is the visualization of the losses during the training (using 50 epochs in that example).

For more details on the meaning of the evaluation metrics, please see Prados et al. (look for: Validation Metrics).