search

SimJeg / FC-DenseNet

Fully Convolutional DenseNets for semantic segmentation.
489 stars 146 forks source link

Error in paper? Figure 1 #28

Open Faur opened 6 years ago

Faur commented 6 years ago

I am wondering if there isn't an error in Figure 1. In order to get m=256 in the last conv. layer there must be a skip connection around the last dense block in the upsampling path, which is valid, since there aren't any transition up blocks

image

decrispell commented 6 years ago

I believe you are right - See Issues #9 and #19.

Faur commented 6 years ago

Yes, thank you!

pjohh commented 6 years ago

According to your code and my understanding of the functionality of a dense block, there should also be a concatenation "around" the bottleneck dense block (concatenate input and output of bottleneck), otherwise number of feature maps dosen't match ... Am I right?