Open codecolony opened 7 years ago
After digging a bit, I think the issue is with data format. I'm using grayscale ground truth. This explains why one dimension is missing? Maybe I should use color images.
Hello @codecolony @ahundt ,
I want to train FC-Densenet on CamVid too. But I don't know how to start the training. I just begin to learn keras, it's difficult for me. Can you give me some details or the commands you have used? And i really don't know how to use densenet.py. Thanks a ton!
looks like the difference between when you have the batches and when you don't. Keras hides the batches when you create your model, that means you only have width height and channels for images in this case, so you don't have to worry about it, but some backend functions use the batch index, so you have batch height width channels. Make sure your dataset inputs are configured correctly, look at the pascal voc example here and in https://github.com/aurora95/Keras-FCN for an example. Grayscale ground truth is the correct thing to use.
Hi,
Thanks for sharing these new implementations. I tried training fcn densenet with camvid dataset but ended up with following error.
ValueError: Error when checking target: expected reshape_2 to have 4 dimensions, but got array with shape (367, 50176, 12)
Could you please let me know what am I doing wrong here?