xmipp3.protocols.protocol_preprocess.protocol_add_noise module
- class xmipp3.protocols.protocol_preprocess.protocol_add_noise.XmippProtAddNoise(**kwargs)[source]
Bases:
ProtRefine3DGiven a sets of volumes or particles the protocol adds noise to them The types of noise are Uniform, Student and Gaussian.
- GAUSSIAN_NOISE = 0
- STUDENT_NOISE = 1
- UNIFORM_NOISE = 2
- class xmipp3.protocols.protocol_preprocess.protocol_add_noise.XmippProtAddNoiseParticles(**kwargs)[source]
Bases:
XmippProtAddNoiseGiven a set of particles, the protocol will add noise to them The types of noise are Uniform, Student and Gaussian.
AI Generated
## Overview
The Add Noise Particles protocol creates a noisy version of an input particle set.
This protocol is useful for simulation, robustness testing, algorithm development, and validation. By adding controlled noise to particle images, the user can test how downstream steps such as 2D classification, alignment, screening, reconstruction, or neural-network methods behave under degraded signal-to-noise conditions.
The protocol supports three noise models:
Gaussian noise;
Student noise;
Uniform noise.
The main output is a new particle set whose images contain the added noise.
## Inputs and General Workflow
The input is a set of particles.
The protocol first writes the input particle set to Xmipp metadata format. Then it runs the Xmipp noise-addition program on the particle metadata and saves the noisy particles into a new stack.
Finally, it creates an output particle set by copying the input particle metadata and updating the image locations so that each particle points to its corresponding noisy image.
## Input Particles
The Input particles parameter defines the particle set to which noise will be added.
The protocol does not change the particle alignment, CTF metadata, sampling rate, or other particle information. It only creates new noisy images and updates the particle image locations.
This makes the output suitable for controlled tests where the metadata should remain the same while the image content becomes noisier.
## Noise Type
The Noise Type parameter selects the statistical distribution of the added noise.
The available options are:
Gaussian adds normally distributed noise.
Student adds noise following a Student distribution.
Uniform adds uniformly distributed noise.
The choice depends on the kind of perturbation the user wants to simulate.
## Gaussian Noise
When Gaussian noise is selected, the user can define:
Standard Deviation;
Mean.
The standard deviation controls the noise strength. Larger values make the particles noisier.
The mean controls the average value of the noise. The default is 0, meaning that the noise does not introduce a systematic intensity offset.
Gaussian noise is the most common option for general signal-to-noise robustness tests.
## Student Noise
When Student noise is selected, the user can define:
Degree of Freedom;
Standard Deviation;
Mean.
The Student distribution can have heavier tails than the Gaussian distribution. This means that extreme noise values may occur more often, especially for low degrees of freedom.
This option is useful when testing algorithms against outlier-prone or non-Gaussian noise.
## Uniform Noise
When Uniform noise is selected, the user can define:
Minimum Value;
Maximum Value.
The protocol samples noise values uniformly between these two limits.
Uniform noise is useful when the user wants noise values bounded within a specific interval.
## Output Particles
The main output is outputParticles.
This output particle set copies the input particle information and replaces the image locations with the generated noisy images.
The noisy particle images are stored in a new stack, and the metadata file created by Xmipp is used to update the particle locations.
The output can be used in downstream protocols like any other particle set.
## Preserved Metadata
The output particles preserve the metadata of the input particles.
This means that alignment parameters, CTF information, sampling rate, and other stored attributes remain available, while the underlying image data are changed to the noisy versions.
This is useful for experiments where the user wants to isolate the effect of image noise while keeping all other processing information unchanged.
## Interpreting the Result
The output particles should be interpreted as synthetic noisy versions of the input particles.
The protocol does not simulate full microscope image formation, radiation damage, CTF effects, detector effects, or realistic background correlations. It only adds random noise according to the selected distribution.
Therefore, the protocol is best understood as a controlled perturbation tool rather than a complete physical simulator.
## Practical Recommendations
Use this protocol when testing the robustness of classification, alignment, screening, or reconstruction protocols.
Start with modest noise levels and increase them gradually.
Use Gaussian noise for general tests, Student noise for heavier-tailed perturbations, and Uniform noise for bounded noise.
Inspect a representative subset of noisy particles before running expensive downstream processing.
Use the same input metadata and compare downstream results between the original and noisy particle sets to assess sensitivity to noise.
Keep the original particle set unchanged and use the output only for testing, simulation, or validation.
## Final Perspective
Add Noise Particles is a synthetic-data utility for perturbing particle images with controlled random noise.
For biological users and method developers, its main value is that it allows controlled experiments on how particle-processing workflows behave as image noise increases.
The protocol is simple by design: it preserves the particle metadata and generates new noisy image data, making it useful for reproducible comparisons between clean and degraded particle sets.
- class xmipp3.protocols.protocol_preprocess.protocol_add_noise.XmippProtAddNoiseVolumes(**kwargs)[source]
Bases:
XmippProtAddNoiseGiven a set of volumes, or a volume the protocol will add noise to them The types of noise are Uniform, Student and Gaussian.
AI Generated
## Overview
The Add Noise Volume/s protocol creates noisy versions of one or more input volumes.
This protocol is useful for simulation, robustness testing, method development, teaching, and validation workflows. By adding controlled noise to a map, the user can test how downstream protocols behave when the signal is degraded.
The protocol supports three types of noise:
Gaussian noise;
Student noise;
Uniform noise.
The output is either a single noisy volume or a set of noisy volumes, depending on whether the input is one volume or a set of volumes.
## Inputs and General Workflow
The input can be either a single volume or a set of volumes.
For each input volume, the protocol runs the Xmipp noise-addition program with the selected noise model and parameters. The noisy map is written to a new MRC file whose name is derived from the input file name and ends with _Noisy.mrc.
The output volume or volume set copies the input metadata and points to the new noisy files.
## Input Volume/s
The Input Volume/s parameter defines the map or maps to which noise will be added.
If the input is a single volume, the protocol creates one noisy output volume.
If the input is a set of volumes, the protocol creates a new set containing one noisy version of each input volume.
The protocol does not align, filter, normalize, or otherwise modify the maps except for adding the selected noise.
## Noise Type
The Noise Type parameter selects the statistical distribution of the noise.
The available options are:
Gaussian adds normally distributed noise.
Student adds noise following a Student distribution.
Uniform adds uniformly distributed noise between a minimum and maximum value.
The choice depends on the simulation or testing scenario. Gaussian noise is the most common general-purpose option. Student noise can generate heavier-tailed perturbations. Uniform noise is useful when the user wants bounded random values.
## Gaussian Noise
When Gaussian noise is selected, the user can define:
Standard Deviation;
Mean.
The standard deviation controls the amplitude of the added noise. Larger values produce noisier maps.
The mean defines the central value of the Gaussian distribution. The default mean is 0, meaning that the noise has no average offset.
Gaussian noise is useful for testing how robust downstream processing is to random additive fluctuations.
## Student Noise
When Student noise is selected, the user can define:
Degree of Freedom;
Standard Deviation;
Mean.
The degree of freedom controls the shape of the Student distribution. Low values can produce heavier tails, meaning that stronger outlier noise values may occur more often than in a Gaussian distribution.
This option is useful when the user wants to test behavior under more outlier-prone noise.
## Uniform Noise
When Uniform noise is selected, the user can define:
Minimum Value;
Maximum Value.
Noise values are sampled uniformly between these two limits.
Uniform noise is useful when the user wants bounded perturbations rather than a distribution with long tails.
## Output Volume
If the input is a single volume, the main output is outputVolume.
This output is the noisy version of the input volume. It copies the input volume information and points to the generated _Noisy.mrc file.
The output can be used in downstream protocols exactly like any other volume.
## Output Volume Set
If the input is a set of volumes, the protocol creates an output volume set.
Each output item corresponds to one input volume, but its file name is replaced by the generated noisy file.
The output set preserves the input set information while replacing the volume locations with the noisy versions.
## Interpreting the Result
The output should be interpreted as the same input map with additive random noise.
No biological or structural information is added. The protocol only degrades or perturbs the input volume according to the selected noise distribution.
Because the result depends on random noise generation, different runs may produce different noisy maps unless the underlying program uses a fixed random seed.
## Practical Recommendations
Use Gaussian noise for standard robustness tests.
Use Student noise when you want heavier-tailed noise with occasional stronger perturbations.
Use Uniform noise when you want bounded random perturbations.
Start with small noise amplitudes and increase them gradually when testing downstream stability.
Inspect the noisy output visually before using it in later analysis.
Keep the original input volumes unchanged and use the noisy outputs only for simulation or validation workflows.
## Final Perspective
Add Noise Volume/s is a simple simulation protocol for perturbing 3D maps.
For biological users and method developers, its main value is that it provides controlled noisy inputs for testing reconstruction, filtering, sharpening, classification, validation, or visualization workflows.
The protocol should be understood as a synthetic-data utility. It does not model the full physical process of cryo-EM image formation; it only adds statistical noise to existing volumes.