Table Detection and Extraction -- TableNet, Deep Learning model with PyTorch from images

The loss function that will be used for this model is torch.nn.BCEWithLogitsLoss() this loss function is a combination of the Sigmoid and the Binary Cross Entropy Loss functions, you can read more about it here. The train function returns a metric dictionary containing the F1 Score, Accuracy, Precision, Recall, and Loss for the current epoch. Note that F1 Score as I said takes into account the recall and precision but I wanted to know which one of these is better or worse. The test function is very similar to the train function and returns the F1 Score, Accuracy, Precision, Recall, and Loss for the current epoch. The model is trained for about 100 epochs with early stopping. In each epoch, I use both the train_on_epoch and the test_on_epoch functions, display them, and check them against the last epoch scores.