gistfile1.txt

import os
import numpy as np
from scipy.misc import imread, imresize
import hickle as hkl
import numpy as np
from kitti_settings import *

#Optput image size
desired_im_sz = (128, 160)

# Create image datasets.
# For one set in GTA we have 2500 images

im_files = []
im_files =  sorted(os.listdir(DATA_DIR))
im_tr_list = im_files[:int(len(im_files)*0.80)]
im_te_list = im_files[int(len(im_files)*0.80):int(len(im_files)*0.95)]
im_ve_list =  im_files[int(len(im_files)*0.95):int(len(im_files)*1)]

source_train  = list(np.full((int(len(im_files)*0.80)), DATA_DIR))
source_test   = list(np.full((int(len(im_files)*0.15)), DATA_DIR))
source_val   = list(np.full((int(len(im_files)*0.05)), DATA_DIR))

def process_data():

    ### Training Data Processing ###
    print "Processing " + str(int(len(im_files)*0.80)) + " training images"
    X = np.zeros((len(im_tr_list),) + desired_im_sz + (3,), np.uint8)
    for i, im_file in enumerate(im_tr_list):
        im = imread(DATA_DIR + im_file)
        X[i] = process_im(im, desired_im_sz)
        
    hkl.dump(X,os.path.join(DATA_DIR, 'X_' + 'train.hkl'))
    hkl.dump(source_train, os.path.join(DATA_DIR, 'sources_train.hkl'))

    ### Testing Data Processing ###
    print "Processing " + str(int(len(im_files)*0.15)) + " testing images"
    X = np.zeros((len(im_te_list),) + desired_im_sz + (3,), np.uint8)
    for i, im_file in enumerate(im_te_list):
        im = imread(DATA_DIR + im_file)
        X[i] = process_im(im, desired_im_sz)
        
    hkl.dump(X,os.path.join(DATA_DIR, 'X_' + 'test.hkl'))
    hkl.dump(source_test, os.path.join(DATA_DIR, 'sources_test.hkl'))

    ### Validation Data Processing ###
    print "Processing " + str(int(len(im_files)*0.05)) + " Validation images"
    X = np.zeros((len(im_ve_list),) + desired_im_sz + (3,), np.uint8)
    for i, im_file in enumerate(im_ve_list):
        im = imread(DATA_DIR + im_file)
        X[i] = process_im(im, desired_im_sz)
        
    hkl.dump(X,os.path.join(DATA_DIR, 'X_' + 'val.hkl'))
    hkl.dump(source_val, os.path.join(DATA_DIR, 'sources_val.hkl'))
# resize and crop image
def process_im(im, desired_sz):
    target_ds = float(desired_sz[0])/im.shape[0]
    im = imresize(im, (desired_sz[0], int(np.round(target_ds * im.shape[1]))))
    d = int((im.shape[1] - desired_sz[1]) / 2)
    im = im[:, d:d+desired_sz[1]]
    return im

if __name__ == '__main__':

    process_data()