Using unet_learner on tiff volume


(Antonio Rueda Toicen) #1

Hi guys,

I want to test unet_learner against another U-net implementation that I’ve built.

I’m scratching my head on how to load the tiff volume and label files that I have.

Here are the files and the paths that I use to load them using imageio.

train_volume_path = 'https://github.com/andandandand/images-for-colab-notebooks/blob/master/train-volume.tif?raw=true'
train_labels_path = 'https://github.com/andandandand/images-for-colab-notebooks/blob/master/train-labels.tif?raw=true'

Here’s an attempt of what I’ve tried: creating a DataBunch out of two DataLoader and my own SegmentationDataset. It’s not working. Any advice? I’m going through the camvid tutorial while I write this.

https://colab.research.google.com/drive/1QW78ohXh764gQnJvyT_Ve4DCib2SLmaE#scrollTo=NJ9LQ_NOYLWH


#2

If you want to use the Dataset class with fastai, you should check this (you’re missing the c attribute here). You really should be creating your custom ItemList and the data block API to benefit from everything the library has to offer.


(Antonio Rueda Toicen) #3

Thanks for the reply! I’ve been reading on the data block API and it solves several problems that I had.