def run_prog():
parser = OptionParser()
parser.add_option("-m", "--model-file", action="store", default="svm_model.pkl")
parser.add_option("-s", "--signed", action="store_true", default=False)
parser.add_option("-r", "--random-window", action="store_true", default=False)
parser.add_option("-n", "--num-scales", action="store", type=int, default=1)
(options, args) = parser.parse_args()
print "Loading model from file ", options.model_file
model_file = open(options.model_file, "rb")
scaler, svc = pickle.load(model_file)
model_file.close()
print "Model loaded"
p_svc = pickle.dumps(svc)
p_scaler = pickle.dumps(scaler)
#imgin = Image.open("images/pedestrian1.png")
imgin = Image.open(args[0])
#imgin = imgin.resize((imgin.size[0] // 2, imgin.size[1] // 2), Image.CUBIC)
humans_scale = []
imgin = imgin.convert("L") # convert to greyscale (luminance)
if options.num_scales == 1:
scales = [1]
else:
scales = np.linspace(1, 0.2, num=5)
if False:
for i in range(len(scales)):
scale = scales[i]
print "Doing scale ", scale, i, "of", len(scales)
img = imgin.resize((int(imgin.size[0] * scale), int(imgin.size[1] * scale)), Image.LINEAR)
img = np.array(img, dtype='d', order='C')
if options.random_window:
img = randWindowExtractor(img, options.signed)
print "Image shape", img.shape
if True:
#img = imgin.resize((int(imgin.size[0] * 0.25), int(imgin.size[1] * 0.25)), Image.LINEAR)
#img = np.array(imgin, dtype='d', order='C')
img = np.array(imgin)
window_hits, detected_humans = detect_humans_multi(img, p_svc, p_scaler, options.signed, debug=True)
#window_hits, detected_humans = detect_humans(img, svc, scaler, options.signed, debug=True)
humans_scale.append((1.0, detected_humans))
fig = plt.figure()
ax = fig.add_subplot(111)
plt.set_cmap(plt.cm.gray)
ax.imshow(np.array(imgin))
#ax.add_patch(matplotlib.patches.Rectangle(((window_hits[0] * cellsize) + window_pixel_shape[0], (window_hits[1] * cellsize) + window_pixel_shape[1]), window_pixel_shape[0], window_pixel_shape[1]))
#for x_win_hits in range(window_hits.shape[0]):
# for y_win_hits in range(window_hits.shape[1]):
# #ax.add_patch(matplotlib.patches.Rectangle((y_win_hits * cellsize, x_win_hits * cellsize), window_pixel_shape[1], window_pixel_shape[0], ec='blue', facecolor='none'))
# if window_hits[x_win_hits, y_win_hits] >= 0.5:
# print (x_win_hits * cellsize), (y_win_hits * cellsize)
# ax.add_patch(matplotlib.patches.Rectangle((y_win_hits * cellsize, x_win_hits * cellsize), window_pixel_shape[1], window_pixel_shape[0], ec='red', facecolor='none', hatch="/"))
for scale, detected_humans in humans_scale:
for x, y, _ in detected_humans:
ax.add_patch(matplotlib.patches.Rectangle((y, x), window_pixel_shape[1] / scale, window_pixel_shape[0] / scale, ec='red', facecolor='none'))
#plt.figure()
#plt.imshow(window_hits)
plt.show()
if len(args) < 2:
raise "Must provide image directory and file to save to"
imgs = DirectoryImagesLoader(args[0])
hogs = []
for i in range(len(imgs)):
print "Running HOG on image", i, "of", len(imgs)
img = imgs.get_image(i)
if options.random_window:
#for i in range(10):
win = randWindowExtractor(img)
_, normcells = HOG.HOG(win, options.signed)
hogs.append(normcells.flatten())
else:
_, normcells = HOG.HOG(img, options.signed)
hogs.append(normcells.flatten())
#sPickle.s_dump_elt(normcells, f)
f = open(args[1], 'wb')
pickle.dump(hogs, f)
f.close()
print "Finished"
