1, if the batch is too small, the training time is not easy convergence, loss easy to shock,
2, may be provided a Freeze layers, not updated parameters
for layer in model.layers:
layer.trainable = Flase
1, if the batch is too small, the training time is not easy convergence, loss easy to shock,
2, may be provided a Freeze layers, not updated parameters
for layer in model.layers:
layer.trainable = Flase