read_data.py

import numpy as np
import cv2
import os
import glob
import pickle
from sklearn import svm
from sklearn.externals import joblib

W = 64
H = 64

hog = cv2.HOGDescriptor((W, H), (16, 16), (8,8), (8,8), 9)
curr_path = os.getcwd()

pos_dir_path = curr_path+'/data/INRIAPerson/train_64x128_H96/pos'
neg_dir_path = curr_path+'/data/INRIAPerson/Train/neg'

pos_img_path = glob.glob(pos_dir_path+'/*.png')
neg_img_path = glob.glob(neg_dir_path+'/*.png')
neg_img_path += glob.glob(neg_dir_path+'/*.jpg')
img = cv2.imread(pos_img_path[1], 2)

train_data = []
train_label = []

for i in range(len(pos_img_path)):
    img = cv2.imread(pos_img_path[i], 2)
    # cv2.imshow('a', img)
    # cv2.waitKey()
    img2 = img[30:30+H, 30:30+W]
    # cv2.imshow('a', img2)
    # cv2.waitKey()
    vec = hog.compute(img2)
    train_data.append(vec.flatten())
    train_label.append(1)
    print("{}: {}".format(i, len(vec)))

with open('POS_data', 'wb') as fp:
    pickle.dump((train_data, train_label), fp)

train_data = []
train_label = []

Test_data = []
Test_label = []

padding = 20

for i in range(len(neg_img_path)):
    img = cv2.imread(neg_img_path[i], 2)
    if img==None:
        continue
    h2, w2 = img.shape
    h2, w2 = int((h2-H)/padding), int((w2-W)/padding)
    for j in range(h2):
        for k in range(w2):
            x = j*padding
            y = k*padding
            img2 = img[x:x+H, y:y+W]
            vec = hog.compute(img2)
            Test_data.append(vec.flatten())
            Test_label.append(0)
            print("{}.{}.{}: {}".format(i, j, k, len(vec)))

with open('NEG_data', 'wb') as fp:
    pickle.dump((Test_data, Test_label), fp)