Closed ahundt closed 7 years ago
Not yet. I just implemented the model from the paper. I still have to see if the implementation is currently correct and then train it on some segmentation dataset.
Cool, thanks for the update. I'm curious if simply tweaking your other densenet implementation to use atrous convolutions could be more effective than the method in the tiramisu paper. I'm planning on doing this with resnet as detailed in this keras-resnet issue first then might try it on densenets as well. Also there are some links to code in that issue that can load the camvid dataset which may save you time.
Have you trained and tested with this? I don't see anything to load a semantic segmentation dataset.