question about 'softmax' in train_vgg.py

[Vincentyao1995]

Hello there!

Your code is pretty clear and helpful to others, thank you for your effort!

Here, I raised a question about 'softmax', u write about this:

train_vgg.py:

#fc8
    with tf.name_scope('fc8') as scope:
        kernel = weight_variable([4096, 2])
        biases = bias_variable([2])
        output_fc8 = tf.nn.relu(fc(output_fc7, kernel, biases), name=scope)

    finaloutput = tf.nn.softmax(output_fc8, name="softmax")

    cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=finaloutput, labels=y))

Here, u used tf.nn.softmax_cross_entropy_with_logits() function, which I thought is inner implemented softmax to normalize data and then calcuate cross entropy, applied on finaloutput which has already been applied by 'softmax'. So is there more reasonable to use:

cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=output_fc8, labels=y))

?

That's to say, use output_fc8 result to calculate cross entropy.

[DirtoryYang]

yes.thanks, It`s so good!

[wy9933]

Thanks, I have the same question