Keras Deep Learning Cookbook

Data processing

Cervical images are of varying sizes and have a high resolution. For CNNs, the incoming data needs to be of uniform size and also needs to have enough resolution to be able to differentiate the main features in classification, but a low enough resolution to avoid computational limits:

# process cervical datasetdef processCervicalData():    # image resizing    imgPaths = []    labels = []    trainingDirs = ['/deeplearning-keras/ch05/data/train']    for dir in trainingDirs:        newFilePaths, newLabels, numLabels = readFilePaths(dir)        if len(newFilePaths) > 0:            imgPaths += newFilePaths            labels += newLabels    imgPaths, labels = shuffle(imgPaths, labels)    labelCount = labelsCount(labels)    type1Count = labelCount[0]    type2Count = labelCount[1] type3Count ...