def process_frame(
frame: numpy.ndarray,
width: int = None,
height: int = None,
mode: str = "RGB",
) -> numpy.ndarray:
"""
Use PIL to resize an RGB frame to a specified height and width \
or changing it to a different mode.
Args:
frame: Target numpy array representing the image that will be resized.
width: Width of the resized image.
height: Height of the resized image.
mode: Passed to Image.convert.
Returns:
The resized frame that matches the provided width and height.
"""
height = height or frame.shape[0]
width = width or frame.shape[1]
frame = Image.fromarray(frame)
frame = frame.convert(mode).resize(size=(width, height))
return numpy.array(frame)
def visualize_image(
image_name: str,
image: np.ndarray,
visualization_dir: pathlib.Path,
targets: List[inference_types.Target],
clf_tiles: List[Tuple[int, int]],
) -> None:
"""Function used to draw boxes and information onto image for visualizing the output
of inference.
Args:
image_name: The original image name used for saving the visualization.
image: The image array.
visualization_dir: Where to save the visualizations.
targets: A list of the targets that were found during inference.
clf_tiles: Which tiles were classified as having targets in them.
"""
# Create a PIL drawable object.
image = Image.fromarray(image).convert("RGBA")
image_draw = ImageDraw.Draw(image)
# Draw the target rectangles.
for target in targets:
top_left = (target.x, target.y)
bottom_right = (target.x + target.width, target.y + target.height)
image_draw.rectangle([top_left, bottom_right],
outline=(0, 255, 0),
width=2)
image_draw.text((target.x, target.y - 10), target.shape.name.lower(),
(0, 255, 0))
# Draw an overlay onto the tiles which were classified as having targets.
w, h = config.CROP_SIZE
overlay = Image.new("RGBA", image.size, (0, 0, 0, 0))
draw = ImageDraw.Draw(overlay)
for group in clf_tiles:
x, y = group
draw.rectangle([(x, y), (x + w, y + h)], fill=(0, 0, 0, 40))
image = Image.alpha_composite(image, overlay)
image.convert("RGB").save((visualization_dir / image_name))
def flatten_alpha(img: np.ndarray) -> np.ndarray:
"""paste an image with alpha onta a solid background"""
# convert and prepare PIL images
img = Image.fromarray(img)
im_flat = Image.new("RGB", img.size, BG_COLOR)
# paste using PIL paste
im_flat.paste(img.convert("RGB"), (0, 0), img.split()[3])
# return numpy array version
return np.array(im_flat)
def hue(im: np.ndarray, hue_lower: float, hue_upper: float) -> np.ndarray:
'''
Randomly disturb the hue of the picture, user can use np.random.seed to fix the random behavior.
Args:
im(np.ndarray): Input image.
hue_lower(float): Lower bound of hue.
hue_upper(float): Upper bound of hue.
'''
hue_delta = np.random.uniform(hue_lower, hue_upper)
im = np.array(im.convert('HSV'))
im[:, :, 0] = im[:, :, 0] + hue_delta
im = PIL.Image.fromarray(im, mode='HSV').convert('RGB')
return im
def resize_frame(frame: np.ndarray, height: int, width: int) -> np.ndarray:
"""
Use PIL to resize an RGB frame to an specified height and width.
Args:
frame: Target numpy array representing the image that will be resized.
height: Height of the resized image.
width: Width of the resized image.
Returns:
The resized frame that matches the provided width and height.
"""
frame = Image.fromarray(frame)
frame = frame.convert("L").resize((height, width))
return np.array(frame)[:, :, None]
def resize_frame(frame: numpy.ndarray,
width: int,
height: int,
mode: str = "RGB") -> numpy.ndarray:
"""
Use PIL to resize an RGB frame to an specified height and width.
Args:
frame: Target numpy array representing the image that will be resized.
width: Width of the resized image.
height: Height of the resized image.
mode: Passed to Image.convert.
Returns:
The resized frame that matches the provided width and height.
"""
frame = Image.fromarray(frame)
frame = frame.convert(mode).resize(size=(width, height))
return numpy.array(frame)