I am trying to reimplement YOLOv4 in TensorFlow with tf.keras. It uses a batch size of 64. This consumes a lot of RAM, so to fit the network into memory, they u
tf.keras