Marmelator/cv-analysis-service
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 37 Wiki Activity

Fix image format (RGB/A, float/uint8, [0, 1/255]) issues

Browse Source
This commit is contained in:
Matthias Bisping 2023-01-11 12:17:07 +01:00
parent d8542762e6
commit d5d67cb064
2 changed files with 126 additions and 82 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

56
cv_analysis/utils/display.py
View File

@ -1,38 +1,48 @@
import cv2
import numpy as np
from PIL import Image
from PIL.Image import Image as Image_t
from matplotlib import pyplot as plt
def show_image_cv2(image, maxdim=700):
h, w, c = image.shape
maxhw = max(h, w)
if maxhw > maxdim:
ratio = maxdim / maxhw
h = int(h * ratio)
w = int(w * ratio)
img = cv2.resize(image, (h, w))
cv2.imshow("", img)
cv2.waitKey(0)
cv2.destroyAllWindows()
def show_image_mpl(image):
fig, ax = plt.subplots(1, 1)
fig.set_size_inches(20, 20)
ax.imshow(image, cmap="gray")
plt.show()
def show_image(image, backend="mpl"):
def show_image(image, backend="mpl", **kwargs):
if backend == "mpl":
show_image_mpl(image)
show_image_mpl(image, **kwargs)
elif backend == "cv2":
show_image_cv2(image)
show_image_cv2(image, **kwargs)
elif backend == "pil":
Image.fromarray(image).show()
else:
raise ValueError(f"Unknown backend: {backend}")
def show_image_cv2(image, maxdim=700, **kwargs):
h, w, c = image.shape
maxhw = max(h, w)
if maxhw > maxdim:
ratio = maxdim / maxhw
h = int(h * ratio)
w = int(w * ratio)
img = cv2.resize(image, (h, w))
cv2.imshow("", img)
cv2.waitKey(0)
cv2.destroyAllWindows()
def show_image_mpl(image, **kwargs):
if isinstance(image, Image_t):
# noinspection PyTypeChecker
image = np.array(image)
# noinspection PyArgumentList
assert image.max() <= 255
fig, ax = plt.subplots(1, 1)
fig.set_size_inches(20, 20)
assert image.dtype == np.uint8
ax.imshow(image, cmap="gray")
ax.title.set_text(kwargs.get("title", ""))
plt.show()
def save_image(image, path):
cv2.imwrite(path, image)

152
test/fixtures/page_generation/page.py vendored
View File

@ -64,35 +64,42 @@ from PIL import Image
# )
# transform = A.Compose(
# [
# # brightness and contrast transforms
# A.OneOf(
# [
# A.RandomGamma(p=0.2),
# A.RandomBrightnessContrast(p=0.2, brightness_limit=0.05, contrast_limit=0.05),
# ],
# p=0.5,
# ),
# # color transforms
# A.SomeOf(
# [
# A.ColorJitter(p=1),
# A.RGBShift(p=1, r_shift_limit=0.3, g_shift_limit=0.3, b_shift_limit=0.3),
# A.ChannelShuffle(p=1),
# ],
# p=1.0,
# n=3, # 3 => all
# ),
# # # blurring and sharpening transforms
# # A.OneOf(
# # [
# # A.GaussianBlur(p=0.05),
# # A.MotionBlur(p=0.05, blur_limit=21),
# # A.Sharpen(p=0.05),
# # ],
# # p=0.0,
# # ),
# ]
# )
from cv_analysis.utils.display import show_image
transform = A.Compose(
[
# # brightness and contrast transforms
A.OneOf(
[
A.RandomGamma(p=0.2),
A.RandomBrightnessContrast(p=0.2, brightness_limit=0.05, contrast_limit=0.05),
],
p=0.5,
),
# color transforms
A.SomeOf(
[
A.ColorJitter(p=1),
A.RGBShift(p=1, r_shift_limit=0.3, g_shift_limit=0.3, b_shift_limit=0.3),
A.ChannelShuffle(p=1),
],
p=1.0,
n=3, # 3 => all
),
# # blurring and sharpening transforms
# A.OneOf(
# [
# A.GaussianBlur(p=0.05),
# A.MotionBlur(p=0.05, blur_limit=21),
# A.Sharpen(p=0.05),
# ],
# p=0.0,
# ),
# A.ColorJitter(p=1),
]
)
@ -100,14 +107,19 @@ transform = A.Compose(
Color = Tuple[int, int, int]
@pytest.fixture(params=["portrait", "landscape"])
@pytest.fixture(
params=[
"portrait",
# "landscape",
]
)
def orientation(request):
return request.param
@pytest.fixture(
params=[
30,
# 30,
100,
]
)
@ -130,7 +142,13 @@ def color_name(request):
return request.param
@pytest.fixture(params=["smooth", "coarse", "neutral"])
@pytest.fixture(
params=[
# "smooth",
# "coarse",
"neutral",
]
)
def texture_name(request):
return request.param
@ -140,7 +158,7 @@ def color(color_name):
return {
"brown": (0.5, 0.3, 0.2),
"yellow": (0.5, 0.5, 0.0),
"sepia": (0.4, 0.3, 0.2),
"sepia": (173, 155, 109),
"gray": (0.3, 0.3, 0.3),
"white": (0.0, 0.0, 0.0),
"light_red": (0.5, 0.0, 0.0),
@ -160,31 +178,62 @@ def texture_fn(texture_name, size):
@pytest.fixture
def texture(texture_fn, size, color):
noise = np.random.rand(*size) * 255
noise = texture_fn(noise)
noise = color_shift_noise(noise, color)
noise = zero_out_below_threshold(noise, 0.1)
return Image.fromarray(noise.astype("uint8"))
noise_arr = np.random.rand(*size) * 255
noise_arr = texture_fn(noise_arr)
noise_arr = color_shift_noise(noise_arr, color)
show_image(noise_arr, backend="mpl", title="noise1")
noise_arr = zero_out_below_threshold(noise_arr, 0.4)
assert noise_arr.max() <= 255
show_image(noise_arr, backend="mpl", title="noise2")
noise_img = Image.fromarray(noise_arr)
show_image(noise_img, backend="mpl", title="noise3")
# noinspection PyTypeChecker
assert np.equal(noise_arr, np.array(noise_img)).all()
return noise_img
def color_shift_noise(noise: np.ndarray, color: Color):
"""Creates a 3-tensor from a 2-tensor by stacking the 2-tensor three times weighted by the color tuple."""
assert noise.ndim == 2
assert isinstance(color, tuple)
assert max(color) <= 255
assert noise.max() <= 255
color = np.array(color)
weights = color / color.sum()
alpha_channel = np.zeros(noise.shape)
colored_noise = np.stack([noise * weight for weight in weights] + [alpha_channel], axis=-1)
assert max(weights) <= 1
alpha_channel = np.ones(noise.shape) * 255
colored_noise = np.stack([noise * weight for weight in weights] + [alpha_channel], axis=-1).astype(np.uint8)
assert colored_noise.shape == (*noise.shape, 4)
return colored_noise
def zero_out_below_threshold(texture, threshold):
assert texture.max() <= 255
threshold = int(texture[:, :, 0:3].sum(axis=2).max() * threshold)
# texture[:, :, 3] = 100
threshold_mask = texture[:, :, 0:3].sum(axis=2) >= threshold
texture[~threshold_mask] = [0, 0, 0, 0]
return texture
@pytest.fixture
def size(dpi, orientation):
if orientation == "portrait":
size = (8.5 * dpi, 11 * dpi)
else:
elif orientation == "landscape":
size = (11 * dpi, 8.5 * dpi)
else:
raise ValueError(f"Unknown orientation: {orientation}")
size = tuple(map(int, size))
return size
@ -192,33 +241,18 @@ def size(dpi, orientation):
@pytest.fixture
def blank_page(size, texture) -> np.ndarray:
"""Creates a blank page with a given orientation and dpi."""
page = Image.fromarray(np.ones((*size, 3), dtype="uint8") * 255)
page = Image.fromarray(np.zeros((*size, 4), dtype=np.uint8) * 255)
page = superimpose_texture_with_transparency(page, texture)
page = np.array(page)
page = transform(image=page)["image"]
show_image(page, backend="mpl", title="page")
# page = transform(image=page)["image"]
return page
def superimpose_texture_with_transparency(page: Image, texture: Image) -> Image:
"""Superimposes a noise image with transparency onto a page image."""
assert page.mode == "RGB"
assert page.mode == "RGBA"
assert texture.mode == "RGBA"
assert page.size == texture.size
texture = texture.convert("RGBA")
show_image(texture, backend="mpl", title="texture")
page.paste(texture, (0, 0), texture)
return page
def zero_out_below_threshold(texture, threshold):
threshold = int(255 * threshold)
texture[:, :, 0] = (texture[:, :, 0] / texture[:, :, 0].max()) * 255
texture[:, :, 1] = (texture[:, :, 1] / texture[:, :, 1].max()) * 255
texture[:, :, 2] = (texture[:, :, 2] / texture[:, :, 2].max()) * 255
texture[:, :, 3] = 25
threshold_mask = texture[:, :, 0] >= threshold
texture[~threshold_mask] = [0, 0, 0, 0]
return texture