Tensorflow-SegNet icon indicating copy to clipboard operation
Tensorflow-SegNet copied to clipboard

Published 20 hours ago • verified publisher icon tkuanlun350

Skip connections?

Open joyousrabbit opened this issue 7 years ago • 2 comments

Hello, it's great work showing how to use conv_transpose. Do you think the skip connections between encoder layer and decoder layer help? Some papers use it others not (your case). What's your opinion?

joyousrabbit avatar Jun 20 '17 21:06 joyousrabbit

I think it is helpful regarding to training speed(converge faster). In my experiment its didn't improve overall performance a lot. Maybe you can use unpooling instead? although current tensorflow doesn't have GPU support for such operation :(

tkuanlun350 avatar Jul 13 '17 05:07 tkuanlun350

Me too I also find that skip connections don't help. It's unresponsible for the papers talk about it. Just a waste of our time.

joyousrabbit avatar Jul 13 '17 09:07 joyousrabbit