search

NifTK / NiftyNet

[unmaintained] An open-source convolutional neural networks platform for research in medical image analysis and image-guided therapy
http://niftynet.io
Apache License 2.0
1.37k stars 404 forks source link

Self-generated data split file doesn't work properly #333

Open kwang-greenhand opened 5 years ago

kwang-greenhand commented 5 years ago

I was trying to use self-generated data split files to define train/valid/inference set, however, I had some very weird problems. The toolbox cannot find the subjects whose ID is smaller than 100. Let's say I have in total 174 subjects, but in the dataset_split file I only included 70, among which 43 has ID larger than 100.

  1. When loading the dataset, the number of subjects is displayed correctly, as well as data partitioning;
  2. Image loader starts to have problems, only 43 were loaded: image In order to figure out which were missing, I switched ACTION to inference, and inferred all training sets, the "inferred.csv" looks like this (see attached files).

I created the datasplit csv file with matlab function "dlmwrite". The automatically generated csv, the self generated csv as well as the inferred.csv files were attached.

Any thoughts regarding this problem? Thanks a lot!!!

dataset_split_problem.zip

kwang-greenhand commented 5 years ago

Follow-up:

Even more weird stuff: This time I didn't make my new csv files. Instead, from the csv generated by the toolbox, I modified it, assigned first 1/10 subjects as inference and left as training, then did inference with "training". Still, the toolbox can only find IDs larger than 100!

I'm totally confused and out of solution. I guess there is a bug here.

NicoYuCN commented 5 years ago

Hi, how is going with this problem? Thanks.