cross_val_splits.py 文件源码

def unet_cross_val(data_dir, out_dir, mapping, splits, unet_conf):

    # Load spreadsheet
    with pd.ExcelFile(mapping) as xls:
        df = pd.read_excel(xls, 'Sheet1').set_index('index')
        df['class'] = df['class'].map({'preplus': 'pre-plus', 'normal': 'normal', 'plus': 'plus'})

    img_dir = join(data_dir, 'images')
    seg_dir = join(data_dir, 'manual_segmentations')
    mask_dir = join(data_dir, 'masks')

    # Check whether all images exist
    check_images_exist(df, img_dir, seg_dir, mask_dir)

    # Now split into training and testing
    CVFile = sio.loadmat(splits)

    # # Combining Pre-Plus and Plus
    # trainPlusIndex = CVFile['trainPlusIndex'][0]
    # testPlusIndex = CVFile['testPlusIndex'][0]
    #
    # plus_dir = make_sub_dir(out_dir, 'trainTestPlus')
    # print "Generating splits for combined No and Pre-Plus"
    # generate_splits(trainPlusIndex, testPlusIndex, df, img_dir, mask_dir, seg_dir, plus_dir)

    # Combining No and Pre-Plus
    trainPrePIndex = CVFile['trainPrePIndex'][0]
    testPrePIndex = CVFile['testPrePIndex'][0]

    prep_dir = make_sub_dir(out_dir, 'trainTestPreP')
    print "Generating splits for combined Pre-Plus and Plus"
    generate_splits(trainPrePIndex, testPrePIndex, df, img_dir, mask_dir, seg_dir, prep_dir)

    # Train models
    train_and_test(prep_dir, unet_conf, processes=1)
    # train_and_test(plus_dir, unet_conf, processes=2)