Computer Vision News - August 2021

2 Summary Co puter Vision Tool 12 The structure expected by InnerEye is as follows: • Each subject has one or more scans, and one or more segmentation masks. There should be one segmentation mask for each ground truth structure (anatomical structure that the model should segment) • For convenience, scans and ground truth masks for different subjects can live in separate folders, but that's not a must. • Inside the root folder for thedataset, there shouldbe afiledataset.csv, containing the following fields at minimum: subject (unique int > 0), channel (scans/ground truth masks), filePath (either nifty, numpy or hdf5 formats). An example would be: dataset_folder_name ├──da taset.csv ├──su bjectID1 / │ ├── ct.nii.gz │ ├── heart.ni i.gz │ └─ ─ lung.nii. gz ├──s u bjectID2/ | ├── ct.nii.gz | ├── heart.ni i.gz | ├── lung.nii. gz ├──... The dataset.csv for this dataset would look like: subject,filePa th,chann el 1,subjectID1 /ct.nii.gz ,ct 1,subjectID1 /heart.ni i.gz, structure1 1,subjectID1 /lung.nii. gz,s tructure2 2,subjectID2 /ct.nii.gz,ct 2,subjectID2 /heart.nii.gz , structure1 2,subjectID2 /lung.nii. gz,s tructure2