def test_transcoding(self):
dct_y, dct_c, dct_r = load(self.jpeg_file_411)
self.assertEqual(dct_y.shape, (50, 75, 64), "wrong dct shape")
self.assertEqual(dct_c.shape, (25, 38, 64), "wrong dct shape")
self.assertEqual(dct_r.shape, (25, 38, 64), "wrong dct shape")
dct_y, dct_c, dct_r = load(self.jpeg_file_420)
self.assertEqual(dct_y.shape, (50, 75, 64), "wrong dct shape")
self.assertEqual(dct_c.shape, (25, 38, 64), "wrong dct shape")
self.assertEqual(dct_r.shape, (25, 38, 64), "wrong dct shape")
dct_y, dct_c, dct_r = load(self.jpeg_file_422)
self.assertEqual(dct_y.shape, (50, 75, 64), "wrong dct shape")
self.assertEqual(dct_c.shape, (25, 38, 64), "wrong dct shape")
self.assertEqual(dct_r.shape, (25, 38, 64), "wrong dct shape")
dct_y, dct_c, dct_r = load(self.jpeg_file_440)
self.assertEqual(dct_y.shape, (50, 75, 64), "wrong dct shape")
self.assertEqual(dct_c.shape, (25, 38, 64), "wrong dct shape")
self.assertEqual(dct_r.shape, (25, 38, 64), "wrong dct shape")
dct_y, dct_c, dct_r = load(self.jpeg_file_444)
self.assertEqual(dct_y.shape, (50, 75, 64), "wrong dct shape")
self.assertEqual(dct_c.shape, (25, 38, 64), "wrong dct shape")
self.assertEqual(dct_r.shape, (25, 38, 64), "wrong dct shape")
def get_image_info(path_to_file):
"""Returns image info as a dictionary with elements, type, width, height, channels"""
assert (fs.file_exists(path_to_file))
img_type = get_image_type(path_to_file)
im = Image.open(path_to_file)
width, height = im.size
channels = im.mode
info_dict = collections.OrderedDict()
info_dict[lookup.file_path] = path_to_file
info_dict[lookup.image_type] = img_type
info_dict[lookup.image_width] = width
info_dict[lookup.image_height] = height
info_dict[lookup.image_channels] = channels
if img_type in lookup.lossy_encoding_types():
coefficients = np.array(load(path_to_file))
ac_dct_y, ac_dct_all = process_coefficient(coefficients)
info_dict[lookup.embedding_max] = ac_dct_y
else:
info_dict[
lookup.
embedding_max] = width * height * convert_channels_to_int(channels)
return info_dict
def test_load(self):
dct_y, dct_c, dct_r = load(self.jpeg_file)
self.assertEqual(dct_y.shape, (205, 205, 64), "wrong dct shape")
self.assertEqual(dct_c.shape, (103, 103, 64), "wrong dct shape")
self.assertEqual(dct_r.shape, (103, 103, 64), "wrong dct shape")
dct_y_nonormalized, dct_c_nonormalized, dct_r_nonormalized = load(
self.jpeg_file, normalized=False)
self.assertEqual(dct_y_nonormalized.shape, (205, 205, 64),
"wrong dct shape")
self.assertEqual(dct_c_nonormalized.shape, (103, 103, 64),
"wrong dct shape")
self.assertEqual(dct_r_nonormalized.shape, (103, 103, 64),
"wrong dct shape")
for c in range(dct_y.shape[-1]):
normalized_range = dct_y.min(), dct_y.max()
unnormalized_range = dct_y_nonormalized.min(
), dct_y_nonormalized.max()
self.assertTrue(
unnormalized_range[0] >= normalized_range[0]
and unnormalized_range[1] <= normalized_range[1],
"normalized shall produce large range of values")
def __getitem__(self, idx):
img, target = self.test_data[idx], self.test_labels[idx]
#img = center_crop(img, (56, 56))
dct_y, dct_cb, dct_cr = load(img)
y_mean, cb_mean, cr_mean = np.load('/home/michal5/cs445/avgs.npy')
y_std, cb_std, cr_std = np.load('/home/michal5/cs445/stds.npy')
dct_cb = upscale(dct_cb)
dct_cr = upscale(dct_cr)
dct_y = np.divide(np.subtract(dct_y, y_mean), y_std)
dct_cb = np.divide(np.subtract(dct_cb, cb_mean), cb_std)
dct_cr = np.divide(np.subtract(dct_cr, cr_mean), cr_std)
dct_y_t = torch.from_numpy(dct_y).float()
dct_cr_t = torch.from_numpy(dct_cr).float()
dct_cb_t = torch.from_numpy(dct_cb).float()
val = torch.cat((dct_y_t, dct_cb_t, dct_cr_t), dim=1)
return val, target
def __getitem__(self, idx):
if self.datatype == 'val':
img, target = self.val_data[idx], self.val_labels[idx]
else:
img, target = self.train_data[idx], self.train_labels[idx]
img = center_crop(img, (320, 320))
#print(img)
dct_y, dct_cb, dct_cr = load(img)
y_mean, cb_mean, cr_mean = np.load(
'/home/michal5/cs445/avgs_imagenette_320.npy')
y_std, cb_std, cr_std = np.load(
'/home/michal5/cs445/stds_imagenette_320.npy')
dct_cb = upscale(dct_cb)
dct_cr = upscale(dct_cr)
dct_y = np.divide(np.subtract(dct_y, y_mean), y_std)
dct_cb = np.divide(np.subtract(dct_cb, cb_mean), cb_std)
dct_cr = np.divide(np.subtract(dct_cr, cr_mean), cr_std)
dct_y_t = torch.from_numpy(dct_y).float()
dct_cr_t = torch.from_numpy(dct_cr).float()
dct_cb_t = torch.from_numpy(dct_cb).float()
val = torch.cat((dct_y_t, dct_cb_t, dct_cr_t), dim=2)
return val, target
def convert_to_dct(img_location):
dct_y, dct_cb, dct_cr = load(img_location)
with open(img_location, 'rb') as src:
buffer = src.read()
dct_y, dct_cb, dct_cr = loads(buffer)
return [dct_y, dct_cb, dct_cr]