Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Features] Support stochastic degradations for RealBasicVSR #647

Merged
merged 3 commits into from
Dec 19, 2021
Merged

[Features] Support stochastic degradations for RealBasicVSR #647

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
d4d6d54
support stochastic degradations
ckkelvinchan Dec 8, 2021
5f9a960
fix bug in RandomBlur
ckkelvinchan Dec 14, 2021
ea54b3a
Fix bug in _apply_random_poisson_noise
ckkelvinchan Dec 14, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
151 changes: 107 additions & 44 deletions mmedit/datasets/pipelines/random_degradations.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,50 +23,76 @@ def __init__(self, params, keys):
self.keys = keys
self.params = params

def get_kernel(self):
def get_kernel(self, num_kernels):
kernel = np.random.choice(
self.params['kernel_list'], p=self.params['kernel_prob'])
kernel_size = random.choice(self.params['kernel_size'])

sigma_x = self.params.get('sigma_x', [0, 0])
sigma_x = np.random.uniform(sigma_x[0], sigma_x[1])
sigma_x_range = self.params.get('sigma_x', [0, 0])
sigma_x = np.random.uniform(sigma_x_range[0], sigma_x_range[1])
sigma_x_step = self.params.get('sigma_x_step', 0)

sigma_y = self.params.get('sigma_y', [0, 0])
sigma_y = np.random.uniform(sigma_y[0], sigma_y[1])
sigma_y_range = self.params.get('sigma_y', [0, 0])
sigma_y = np.random.uniform(sigma_y_range[0], sigma_y_range[1])
sigma_y_step = self.params.get('sigma_y_step', 0)

rotate_angle = self.params.get('rotate_angle', [-np.pi, np.pi])
rotate_angle = np.random.uniform(rotate_angle[0], rotate_angle[1])
rotate_angle_range = self.params.get('rotate_angle', [-np.pi, np.pi])
rotate_angle = np.random.uniform(rotate_angle_range[0],
rotate_angle_range[1])
rotate_angle_step = self.params.get('rotate_angle_step', 0)

beta_gau = self.params.get('beta_gaussian', [0.5, 4])
beta_gau = np.random.uniform(beta_gau[0], beta_gau[1])
beta_gau_range = self.params.get('beta_gaussian', [0.5, 4])
beta_gau = np.random.uniform(beta_gau_range[0], beta_gau_range[1])
beta_gau_step = self.params.get('beta_gaussian_step', 0)

beta_pla = self.params.get('beta_plateau', [1, 2])
beta_pla = np.random.uniform(beta_pla[0], beta_pla[1])
beta_pla_range = self.params.get('beta_plateau', [1, 2])
beta_pla = np.random.uniform(beta_pla_range[0], beta_pla_range[1])
beta_pla_step = self.params.get('beta_plateau_step', 0)

omega = self.params.get('omega', None)
if omega is not None:
omega = np.random.uniform(omega[0], omega[1])
else: # follow Real-ESRGAN
omega_range = self.params.get('omega', None)
omega_step = self.params.get('omega_step', 0)
if omega_range is None: # follow Real-ESRGAN settings if not specified
if kernel_size < 13:
omega = np.random.uniform(np.pi / 3, np.pi)
omega_range = [np.pi / 3., np.pi]
else:
omega = np.random.uniform(np.pi / 5, np.pi)
omega_range = [np.pi / 5., np.pi]
omega = np.random.uniform(omega_range[0], omega_range[1])

# determine blurring kernel
kernel = blur_kernels.random_mixed_kernels(
[kernel],
[1],
kernel_size,
[sigma_x, sigma_x],
[sigma_y, sigma_y],
[rotate_angle, rotate_angle],
[beta_gau, beta_gau],
[beta_pla, beta_pla],
[omega, omega],
None,
)

return kernel
kernels = []
for _ in range(0, num_kernels):
kernel = blur_kernels.random_mixed_kernels(
[kernel],
[1],
kernel_size,
[sigma_x, sigma_x],
[sigma_y, sigma_y],
[rotate_angle, rotate_angle],
[beta_gau, beta_gau],
[beta_pla, beta_pla],
[omega, omega],
None,
)
kernels.append(kernel)

# update kernel parameters
sigma_x += np.random.uniform(-sigma_x_step, sigma_x_step)
sigma_y += np.random.uniform(-sigma_y_step, sigma_y_step)
rotate_angle += np.random.uniform(-rotate_angle_step,
rotate_angle_step)
beta_gau += np.random.uniform(-beta_gau_step, beta_gau_step)
beta_pla += np.random.uniform(-beta_pla_step, beta_pla_step)
omega += np.random.uniform(-omega_step, omega_step)

sigma_x = np.clip(sigma_x, sigma_x_range[0], sigma_x_range[1])
sigma_y = np.clip(sigma_y, sigma_y_range[0], sigma_y_range[1])
rotate_angle = np.clip(rotate_angle, rotate_angle_range[0],
rotate_angle_range[1])
beta_gau = np.clip(beta_gau, beta_gau_range[0], beta_gau_range[1])
beta_pla = np.clip(beta_pla, beta_pla_range[0], beta_pla_range[1])
omega = np.clip(omega, omega_range[0], omega_range[1])

return kernels

def _apply_random_blur(self, imgs):
is_single_image = False
Expand All @@ -75,8 +101,11 @@ def _apply_random_blur(self, imgs):
imgs = [imgs]

# get kernel and blur the input
kernel = self.get_kernel()
imgs = [cv2.filter2D(im, -1, kernel) for im in imgs]
kernels = self.get_kernel(num_kernels=len(imgs))
imgs = [
cv2.filter2D(img, -1, kernel)
for img, kernel in zip(imgs, kernels)
]

if is_single_image:
imgs = imgs[0]
Expand Down Expand Up @@ -136,6 +165,8 @@ def _random_resize(self, imgs):
'implemented')
resize_opt = self.resize_dict[resize_opt]

resize_step = self.params.get('resize_step', 0)

# determine the target size, if not provided
target_size = self.params.get('target_size', None)
if target_size is None:
Expand All @@ -149,16 +180,32 @@ def _random_resize(self, imgs):
else:
scale_factor = 1
target_size = (int(h * scale_factor), int(w * scale_factor))
else:
resize_step = 0

# resize the input
imgs = [
cv2.resize(img, target_size[::-1], interpolation=resize_opt)
for img in imgs
]
if resize_step == 0: # same target_size for all input images
outputs = [
cv2.resize(img, target_size[::-1], interpolation=resize_opt)
for img in imgs
]
else: # different target_size for each input image
outputs = []
for img in imgs:
img = cv2.resize(
img, target_size[::-1], interpolation=resize_opt)
outputs.append(img)

# update scale
scale_factor += np.random.uniform(-resize_step, resize_step)
scale_factor = np.clip(scale_factor, resize_scale[0],
resize_scale[1])
target_size = (int(h * scale_factor), int(w * scale_factor))

if is_single_image:
imgs = imgs[0]
outputs = outputs[0]

return imgs
return outputs

def __call__(self, results):
if np.random.uniform() > self.params.get('prob', 1):
Expand Down Expand Up @@ -197,6 +244,8 @@ def _apply_gaussian_noise(self, imgs):
sigma_range = self.params['gaussian_sigma']
sigma = np.random.uniform(sigma_range[0], sigma_range[1]) / 255.

sigma_step = self.params.get('gaussian_sigma_step', 0)

gray_noise_prob = self.params['gaussian_gray_noise_prob']
is_gray_noise = np.random.uniform() < gray_noise_prob

Expand All @@ -207,12 +256,19 @@ def _apply_gaussian_noise(self, imgs):
noise = noise[:, :, :1]
outputs.append(img + noise)

# update noise level
sigma += np.random.uniform(-sigma_step, sigma_step) / 255.
sigma = np.clip(sigma, sigma_range[0] / 255.,
sigma_range[1] / 255.)

return outputs

def _apply_poisson_noise(self, imgs):
scale_range = self.params['poisson_scale']
scale = np.random.uniform(scale_range[0], scale_range[1])

scale_step = self.params.get('poisson_scale_step', 0)

gray_noise_prob = self.params['poisson_gray_noise_prob']
is_gray_noise = np.random.uniform() < gray_noise_prob

Expand All @@ -228,6 +284,10 @@ def _apply_poisson_noise(self, imgs):

outputs.append(img + noise * scale)

# update noise level
scale += np.random.uniform(-scale_step, scale_step)
scale = np.clip(scale, scale_range[0], scale_range[1])

return outputs

def _apply_random_noise(self, imgs):
Expand Down Expand Up @@ -289,23 +349,26 @@ def _apply_random_compression(self, imgs):
is_single_image = True
imgs = [imgs]

# determine compression level
# determine initial compression level and the step size
quality = self.params['quality']
quality_step = self.params.get('quality_step', 0)
jpeg_param = round(np.random.uniform(quality[0], quality[1]))
encode_param = [int(cv2.IMWRITE_JPEG_QUALITY), jpeg_param]

# apply jpeg compression
outputs = []
for img in imgs:
encode_param = [int(cv2.IMWRITE_JPEG_QUALITY), jpeg_param]
_, img_encoded = cv2.imencode('.jpg', img * 255., encode_param)
outputs.append(np.float32(cv2.imdecode(img_encoded, 1)) / 255.)

imgs = outputs
# update compression level
jpeg_param += np.random.uniform(-quality_step, quality_step)
jpeg_param = round(np.clip(jpeg_param, quality[0], quality[1]))

if is_single_image:
imgs = imgs[0]
outputs = outputs[0]

return imgs
return outputs

def __call__(self, results):
if np.random.uniform() > self.params.get('prob', 1):
Expand Down
18 changes: 15 additions & 3 deletions tests/test_data/test_pipelines/test_random_degradations.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,12 +111,24 @@ def test_random_resize():
resize_scale=[0.5, 1.5],
resize_opt=['bilinear', 'area', 'bicubic'],
resize_prob=[1 / 3., 1 / 3., 1 / 3.],
target_size=(16, 16)),
target_size=(16, 32)),
keys=['lq'])
results = model(results)
assert results['lq'].shape == (16, 16, 3)
assert results['lq'].shape == (16, 32, 3)

# skip degrdation
# step_size > 0
results['lq'] = np.ones((8, 8, 3)).astype(np.float32)
model = RandomResize(
params=dict(
resize_mode_prob=[0, 0, 1],
resize_scale=[0.5, 1.5],
resize_opt=['bilinear', 'area', 'bicubic'],
resize_prob=[1 / 3., 1 / 3., 1 / 3.],
resize_step=0.05),
keys=['lq'])
results = model(results)

# skip degradation
model = RandomResize(
params=dict(
resize_mode_prob=[1, 0, 0],
Expand Down