Focal-Loss-implement-on-Tensorflow icon indicating copy to clipboard operation
Focal-Loss-implement-on-Tensorflow copied to clipboard

Published 20 hours ago • verified publisher icon ailias

tf.nn.sigmoid and tf.nn.softmax diff?

Open ronyuzhang opened this issue 6 years ago • 1 comments

ronyuzhang avatar Mar 12 '18 12:03 ronyuzhang

The paper of He proposed sigmoid has more robust training effect than softmax, But he didn't explain why. You can have a try. I think maybe it depends the dependent of each probability of result, softmax may have more small probability values(all sum is 1), which may cause more negative samples.

ChiefGodMan avatar Mar 14 '18 02:03 ChiefGodMan