search

preddy5 / segnet

A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation
http://preddy5.github.io/2016/03/08/segnet-post.html
205 stars 106 forks source link

Feature Request: indices-based upsampling layer #4

Open waissbluth opened 7 years ago

waissbluth commented 7 years ago

Hello, does anyone have an implementation of this, or any thoughts on how to implement it?

Thanks

preddy5 commented 7 years ago

Hey @waissbluth I didnt get enough time to work on it yet, I'll probably work on in a week or two.

waissbluth commented 7 years ago

Hey, is there anything I can do to help out with this? would really appreciate the feature.

mpariente commented 7 years ago

Hey, I would also love this feature! :) Can I do anything to help?

Maybe a custom Maxpooling2D layer which produces an auxiliary output and an Upsample2D which would take an optionnal additionnal input would work?

redouanelg commented 7 years ago

Hi all, this guy implemented a class he called Depool2D, I'm not sure if it's what you are looking for https://github.com/nanopony/keras-convautoencoder/blob/master/autoencoder_layers.py

mpariente commented 7 years ago

Thanks for the link ! I saw it already and I have troubles understanding his DependentDense. And also his DePool2D doesn't receive a switch from the MaxPooling2D if I read well.