How did you get the count64 files

[findalexli]

Hello folks.

I attended the hackathon as well. I am trying to reproduce the result, but I cannot find how did you generate the count64 files.

count64_file = os.path.join(subject_folder, "count64_file.nii.gz")
count64_vol = nibabel.load(count64_file)
count64_data = count64_vol.get_data()
self.count64_lengths.append(count64_data.size)

[eickenberg]

Hi, I made those by convolving every tumor indicator image with an image full of ones of size 64x64. Because the indicator is made to contain 1 when there is tumor tissue and 0 when not, this convolution effectively counts the tumor pixels for every 64x64 patch. I saved that out as a nifti file. This file may still be available somewhere, but the VM we worked on got shut down, so it might need to be re-created. I think it turns out that the balancing method which uses this file was not actually very important, or potentially even counterproductive. So you might want to just go ahead without it and select uniformly random patches in the dataloader

…

On Sun, Mar 1, 2020 at 10:01 PM alex li ***@***.***> wrote: Hello folks. I attended the hackathon as well. I am trying to reproduce the result, but I cannot find how did you generate the count64 files. count64_file = os.path.join(subject_folder, "count64_file.nii.gz") count64_vol = nibabel.load(count64_file) count64_data = count64_vol.get_data() self.count64_lengths.append(count64_data.size) — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJ7AC3V3LVWZCSHMGA7OMTRFMOPVANCNFSM4K7LIJHQ> .