def crop(img, illumination):
if img is None:
return None
img = img[start_x:start_x + s, start_y:start_y + s]
img = rotate_and_crop(img, angle)
img = cv2.resize(img, (FLAGS.img_size, FLAGS.img_size))
if FLAGS.AUGMENTATION_FLIP_LEFTRIGHT and flip_lr:
img = img[:, ::-1]
if FLAGS.AUGMENTATION_FLIP_TOPDOWN and flip_td:
img = img[::-1, :]
img = img.astype(np.float32)
new_illum = np.zeros_like(illumination)
# RGB -> BGR
illumination = illumination[::-1]
for i in range(3):
for j in range(3):
new_illum[i] += illumination[j] * color_aug[i, j]
if FLAGS.AUGMENTATION_COLOR_OFFDIAG > 0:
# Matrix mul, slower
new_image = np.zeros_like(img)
for i in range(3):
for j in range(3):
new_image[:, :, i] += img[:, :, j] * color_aug[i, j]
else:
img *= np.array(
[[[color_aug[0][0], color_aug[1][1], color_aug[2][2]]]],
dtype=np.float32)
new_image = img
new_image = np.clip(new_image, 0, 65535)
def apply_nonlinearity(image):
if FLAGS.AUGMENTATION_GAMMA != 0 or FLAGS.USE_CURVE:
res = 1024
image = np.clip(image * (res * 1.0 / 65536), 0, res - 1)
gamma = 1.0 + (random.random() -
0.5) * FLAGS.AUGMENTATION_GAMMA
if USE_CURVE:
curve = get_random_curve()
else:
curve = lambda x: x
mapping = create_lut(lambda x: curve(x)**gamma * 65535.0, res)
return mapping(image)
else:
return image
if FLAGS.SPATIALLY_VARIANT:
split = new_image.shape[1] / 2
new_image[:, :split] = apply_nonlinearity(new_image[:, :split])
new_image[:, split:] = apply_nonlinearity(new_image[:, split:])
else:
new_image = apply_nonlinearity(new_image)
new_illum = np.clip(new_illum, 0.01, 100)
return new_image, new_illum[::-1]
def crop(img):
if img is None:
return None
img = img[start_x:start_x + s, start_y:start_y + s]
img = rotate_and_crop(img, angle)
img = cv2.resize(img, (self.input_size[1], self.input_size[0]))
img = img.astype(np.float32)
new_image = img
new_image = np.clip(new_image, 0, 255.0)
#plt.imshow(new_image);plt.show()
return new_image
def crop(img, illumination):
if img is None:
return None
img = img[start_x:start_x + s, start_y:start_y + s]
img = rotate_and_crop(img, angle)
img = cv2.resize(img, (FCN_INPUT_SIZE, FCN_INPUT_SIZE))
if flip_lr:
img = img[:, ::-1]
img = img.astype(np.float32)
new_illum = np.zeros_like(illumination)
# RGB -> BGR
illumination = illumination[::-1]
for i in range(3):
for j in range(3):
new_illum[i] += illumination[j] * color_aug[i, j]
img *= np.array(
[[[color_aug[0][0], color_aug[1][1], color_aug[2][2]]]],
dtype=np.float32)
new_image = img
new_image = np.clip(new_image, 0, 65535)
new_illum = np.clip(new_illum, 0.01, 100)
return new_image, new_illum[::-1]