Denoising

class topaz.denoise.AffineDenoise(max_size=31)
forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class topaz.denoise.AffineFilter(weights)
forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class topaz.denoise.DenoiseNet(base_filters)
forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class topaz.denoise.DenoiseNet2(base_filters, width=11)
forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class topaz.denoise.GaussianDenoise(sigma, scale=5)
forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class topaz.denoise.GaussianDenoise3d(sigma, scale=5)
forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class topaz.denoise.GaussianNoise(x, sigma=1.0, crop=500, xform=True)
class topaz.denoise.Identity
forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class topaz.denoise.InvGaussianFilter(sigma, scale=5)
forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class topaz.denoise.L0Loss(eps=1e-08, gamma=2)
class topaz.denoise.NoiseImages(x, crop=800, xform=True, preload=False, cutoff=0)
load_image(path)
class topaz.denoise.PairedImages(x, y, crop=800, xform=True, preload=False, cutoff=0)
load_image(path)
class topaz.denoise.UDenoiseNet(nf=48, base_width=11, top_width=3)
forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class topaz.denoise.UDenoiseNet2(nf=48)
forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class topaz.denoise.UDenoiseNet3
forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class topaz.denoise.UDenoiseNet3D(nf=48, base_width=11, top_width=3)
forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class topaz.denoise.UDenoiseNetSmall(nf=48, width=11, top_width=3)
forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
topaz.denoise.correct_spatial_covariance(x, width=11, s=11, patch=1)

Estimates the spatial covariance in the micrograph, finds closest Guassian kernel parameters that would give this covariance structure, and applies inverse of that filter to correct the spatial covariance.

topaz.denoise.denoise(model, x, patch_size=- 1, padding=128)
topaz.denoise.denoise_patches(model, x, patch_size, padding=128)
topaz.denoise.denoise_stack(model, stack, batch_size=20, use_cuda=False)
topaz.denoise.estimate_unblur_filter(x, width=11, s=11)

Estimate parameters of the affine filter that would give zero autocovariance to the image.

topaz.denoise.estimate_unblur_filter_gaussian(x, width=11, s=11)

Estimate parameters of the Gaussian filter that would give the closest spatial covariance structure to that observed in image x if the unfiltered image had zero spatial covariance. Then, return inverse of that filter.

topaz.denoise.eval_mask_denoise(model, dataset, criteria, p=0.01, batch_size=10, use_cuda=False, num_workers=0)
topaz.denoise.eval_noise2noise(model, dataset, criteria, batch_size=10, use_cuda=False, num_workers=0)
topaz.denoise.gaussian(x, sigma=1, scale=5, use_cuda=False)

Apply Gaussian filter with sigma to image. Truncates the kernel at scale times sigma pixels

topaz.denoise.gaussian3d(x, sigma=1, scale=5, use_cuda=False)

Apply Gaussian filter with sigma to volume. Truncates the kernel at scale times sigma pixels

topaz.denoise.gaussian_filter(sigma, s=11)
topaz.denoise.gaussian_filter_3d(sigma, s=11)
topaz.denoise.inverse_filter(w)
topaz.denoise.load_model(name)
topaz.denoise.lowpass(x, factor=1)

low pass filter with FFT

topaz.denoise.lowpass3d(x, factor=1)

low pass filter with FFT

topaz.denoise.spatial_covariance(x, n=11, s=11)
topaz.denoise.spatial_covariance_old(x, n=11, s=11)
topaz.denoise.train_mask_denoise(model, dataset, p=0.01, lr=0.001, optim='adagrad', batch_size=10, num_epochs=100, criteria=MSELoss(), dataset_val=None, use_cuda=False, num_workers=0, shuffle=True)
topaz.denoise.train_noise2noise(model, dataset, lr=0.001, optim='adagrad', batch_size=10, num_epochs=100, criteria=MSELoss(), dataset_val=None, use_cuda=False, num_workers=0, shuffle=True)