search

0bserver07 / One-Hundred-Layers-Tiramisu

Keras Implementation of The One Hundred Layers Tiramisu: Fully Convolutional DenseNets for Semantic Segmentation by (Simon Jégou, Michal Drozdzal, David Vazquez, Adriana Romero, Yoshua Bengio)
https://arxiv.org/abs/1611.09326
MIT License
197 stars 54 forks source link

concatenation in denseblock #22

Open jyu-theartofml opened 5 years ago

jyu-theartofml commented 5 years ago

Hi, Maybe I'm not reading your code right, but I don't see a concatenation step after each convolution layer in the denseblock() function. I thought this was needed as shown in the paper.

THanks,

QingboKang commented 5 years ago

Agree with @yinniyu , I didn't see any concatenation operation inside the DenseBlock, if no dense connection, the network almost has no difference with U-Net except has more layers.

lancesye commented 5 years ago

I don't see the skip connection in the denseblock() too. I plot the model. It is only a deep unet