What's the purpose of "ignoring last label" in this loss function?

[lishen]

See code:

Keras-FCN/utils/loss_function.py

Line 14 in 35afe12

y_true = K.one_hot(tf.to_int32(K.flatten(y_true)), K.int_shape(y_pred)[-1]+1)

Keras-FCN/utils/loss_function.py

Line 15 in 35afe12

unpacked = tf.unstack(y_true, axis=-1)

Keras-FCN/utils/loss_function.py

Line 16 in 35afe12

y_true = tf.stack(unpacked[:-1], axis=-1)

At first you appended an additional column of all 0's in one-hot coding and then removed it. It seems to me you only need a single call of K.one_hot to replace L14-16. Or, did I miss something?

[Nandayang]

Hi, I have the same question as yours. Do you have any answer about this question?3KS!@lishen