search

MarvinLer / tcga_segmentation

Whole Slide Image segmentation with weakly supervised multiple instance learning on TCGA | MICCAI2020 https://arxiv.org/abs/2004.05024
GNU Affero General Public License v3.0
124 stars 34 forks source link

Question regarding the data shape #9

Closed manigalati closed 3 years ago

manigalati commented 3 years ago

Hello :) I have a question regarding the dimensions of the data loaded at each epoch from the dataloader. Since the GDC Data Portal seems to be down, I cannot download any data to run an example. My only option is to run your code by plugging my data. However, I don't know the shape each batch of data has at each epoch. My guess is that the shape of the images is 1xBx3xWxH, where B is the max bag size you can set as parameter. Instead, the shape of the label is 1, meaning that all the slides loaded at one epoch has to belong to the same class. Are my assumptions correct?

Thank you in advance :)

Francesco

MarvinLer commented 3 years ago

Hi, Sorry for the late reply, not sure if you figured this out yet

The shape of the images is indeed 1xBx3xWxH with B equal to the --max-bag-size parameter, i.e. the number of tiles sampled per WSI. E.g. a shape [1, 100, 3, 224, 224]. The shape of the label is indeed [1].

Is it okay for you? Did you manage to use it with your own data?

Best, ML