img_preprocess.py

import numpy as np
import errno
import glob
import pickle
import re

def GetCoordinate(data_dir):
    coordinate = []
    path = data_dir + '*.txt'
    files = glob.glob(path)
    for name in files:
        try:
            with open(name) as f:
                cars = []
                for i, t in enumerate(f.readlines(), 1):
                    t = t.rstrip("\n")
                    t = t.replace('\t', ' ')
                    l_s = re.split('\(|\)| |\*',t)
                    dic = {}
                    #Loads the box dimensions from file
                    dic['x1'] = float(l_s[0])
                    dic['y1'] = float(l_s[1])
                    dic['x2'] = float(l_s[0]) + float(l_s[2])
                    dic['y2'] = float(l_s[1]) + float(l_s[3])
                    cars.append(dic)
            coordinate.append(cars)
        except IOError as exc:
            if exc.errno != errno.EISDIR:
                print("error occurred here")
                raise
    return coordinate

def createBoxes(images, coordinates, scaling):
    seq_length = len(coordinates)
    _,a,b = images.shape
    for n in range(seq_length):
        if (n % 100 == 0):
            print("Processing the " + str(n) + "th image\n")
        frame = coordinates[n]
        for i in range(len(frame)):
            #Create a map of zeros and ones inside the box
            car = frame[i]
            pixel_start_x = int(round(float(car['x1'])/scaling))
            pixel_start_y = int(round(float(car['y1'])/scaling))
            pixel_end_x = int(round(float(car['x2'])/scaling))
            pixel_end_y = int(round(float(car['y2'])/scaling))
            for ri in range(pixel_end_y-pixel_start_y+1):
                for ci in range(pixel_end_x-pixel_start_x+1):
                    images[n, pixel_start_y+ri, pixel_start_x+ci] = 1
    print("Processing complete!\n")
    return images

def getScaling(dim):
    scaling = 1
    #Shrinks the image down by a factor of 2 until dimension<32 or odd
    while (dim[0]%2 == 0 and dim[1]%2 == 0 and min(dim)>32):
        dim = (dim[0]/2, dim[1]/2)
        scaling *= 2
    return dim, scaling

def saveImages(images, name):
    pickle.dump(images, open(name + ".pkl", "wb"))

if __name__ == "__main__":
    cc = GetCoordinate('./Car/')
    image_dim = (240, 320)
    image_dim, scaling = getScaling(image_dim)
    print("The image dimension after shrinking is " + str(image_dim) + "\n")
    a = int(image_dim[0])
    b = int(image_dim[1])
    images = np.zeros((6043,a,b))

    new_images = createBoxes(images, cc, scaling)
    import pdb
    pdb.set_trace()
    saveImages(new_images, 'bbox_data')