tf-unet

tensorflow version of unet

u-net is defined in the UNet.py

To make sure you don't consider the pixels with zero labels i.e. pixels with missing labels, do the following

loss_map = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=pred, labels=gt)
loss_map = tf.multiply(loss_map,tf.to_float(tf.not_equal(gt,0)))

to train

python3 train_u-net.py