predict - Predict unnormalized anomaly scores - MATLAB (original) (raw)
Predict unnormalized anomaly scores
Since R2022b
Syntax
Description
[scores](#mw%5Fb7a4cd21-1647-4dd1-9a56-1023f7cff936) = predict([detector](#mw%5F0cc1df5f-fa72-42ff-96f8-fa3591b6588b%5Fsep%5Fmw%5Ff0686b83-345a-47b2-8ed0-7c3d0dde776c),[I](#mw%5F5fcdda5d-8814-409e-9023-b8e2fdc4a762)) calculates the unnormalized anomaly scores predicted by an anomaly detector during inference for a set of test images, I. Use this function to get predictions from the output layers of the detector during inference.
Note
This functionality requires Deep Learning Toolbox™ and the Automated Visual Inspection Library for Computer Vision Toolbox™. You can install the Automated Visual Inspection Library for Computer Vision Toolbox from Add-On Explorer. For more information about installing add-ons, see Get and Manage Add-Ons.
[scores](#mw%5Fb7a4cd21-1647-4dd1-9a56-1023f7cff936) = predict([detector](#mw%5F0cc1df5f-fa72-42ff-96f8-fa3591b6588b%5Fsep%5Fmw%5Ff0686b83-345a-47b2-8ed0-7c3d0dde776c),[I](#mw%5F5fcdda5d-8814-409e-9023-b8e2fdc4a762),[Name=Value](#namevaluepairarguments)) specifies options using one or more name-value arguments. For example,predict(detector,I,MiniBatchSize=32) limits the batch size to 32.
Examples
Load calibration images and corresponding labels, then create a datastore that reads the calibration data. The data set consists of grayscale images of handwritten digits 0–9.
[Xcal,gtLabels] = digitTest4DArrayData; dsCal = arrayDatastore(Xcal,IterationDimension=4);
Load a pretrained FCDD anomaly detector. This detector has been trained to classify the digit 8 as normal and all other digits as anomalies. Therefore, specify the set of anomaly labels as the set of digits between 0 and 9, excluding 8.
load("digit8AnomalyDetector.mat"); anomalyLabels = setdiff(string(0:9),"8");
Predict the anomaly score of each calibration image.
scores = predict(detector,dsCal);
Calculate the optimal threshold and corresponding ROC metrics from the anomaly scores and ground truth labels.
[T,roc] = anomalyThreshold(gtLabels,scores,anomalyLabels)
roc = rocmetrics with properties:
Metrics: [4976×4 table]Properties, Methods
Set the Threshold property of the FCDD anomaly detector as the optimal threshold.
Input Arguments
Test image, specified in one of these formats:
| Format | Supported Detectors |
|---|---|
| _M_-by-_N_-by-3 numeric array representing a truecolor image. | EfficientAD, FCDD, FastFlow, PatchCore |
| _M_-by-_N_-by-3-by-B numeric array representing a batch of B truecolor images. | EfficientAD, FCDD, FastFlow, PatchCore |
| Datastore that reads and returns truecolor images. The images must all have the same size. | EfficientAD, FCDD, FastFlow, PatchCore |
| Formatted dlarray (Deep Learning Toolbox) object with two spatial dimensions and one channel dimension. You can specify multiple test images by including a batch dimension. | EfficientAD, FCDD, FastFlow |
FCDD anomaly detectors also support grayscale test images, with one color channel instead of three.
Name-Value Arguments
Specify optional pairs of arguments asName1=Value1,...,NameN=ValueN, where Name is the argument name and Value is the corresponding value. Name-value arguments must appear after other arguments, but the order of the pairs does not matter.
Example: isAnomaly = predict(detector,I,MiniBatchSize=32) limits the batch size to 32.
Size of batches for calculating predictions, specified as a positive integer. Larger batch sizes lead to faster processing but take up more memory.
Function for computing a scalar score from an anomaly map whendetector is an efficientADAnomalyDetector object, an fcddAnomalyDetector object, or a fastFlowAnomalyDetector object, specified as a function handle. The function handle ScoreFunction must represent a function that accepts the input data I and returns an output in the form of a numeric scalar. I must be 2-D numeric image data, dlarray data with two spatial dimensions, or a datastore. The default value of ScoreFunction depends on the type of detector specified:
| Detector Specified in detector Argument | Default ScoreFunction Value |
|---|---|
| FastFlow | @(I)max(I,[],[1 2]) |
| EfficientAD | @(I)max(I,[],[1 2 3]) |
| FCDD | @(I)mean(I,[1 2]) |
| PatchCore | Not supported |
Hardware resource on which to run the detector, specified as"auto", "gpu", or "cpu". The table shows the valid hardware resource values.
| Resource | Action |
|---|---|
| "auto" | Use a GPU if it is available. Otherwise, use the CPU. |
| "gpu" | Use the GPU. To use a GPU, you must have Parallel Computing Toolbox™ and a CUDA® enabled NVIDIA® GPU. If a suitable GPU is not available, the function returns an error. For information about the supported compute capabilities, seeGPU Computing Requirements (Parallel Computing Toolbox). |
| "cpu" | Use the CPU. |
Output Arguments
Predicted anomaly scores, returned as a numeric scalar, numeric vector, or formatteddlarray (Deep Learning Toolbox) object.scores contains one element for each image inI.
Extended Capabilities
Usage notes and limitations:
| FCDD, FastFlow, PatchCore | EfficientAD |
|---|---|
| The size of I must be fixed at code generation time.The only name-value argument supported for code generation isMiniBatchSize. The name-value argument must be a compile-time constant. | The datastore data format of I is not supported. |
Usage notes and limitations:
| FCDD, FastFlow, PatchCore | EfficientAD |
|---|---|
| The size of I must be fixed at code generation time.GPU code generation does not support gpuArray inputs.GPU code generation supports inputs that are defined as half-precision floating point data types. For more information, see half (GPU Coder).The only name-value argument supported for code generation isMiniBatchSize. The name-value argument must be a compile-time constant. | The datastore data format of I is not supported. |
Version History
Introduced in R2022b
Specify a custom anomaly score function, which the predict function uses to compute a scalar score from the anomaly map, by using the ScoreFunction name-value argument.
predict now supports the generation of C code (requires MATLAB® Coder™) and optimized CUDA code (requires GPU Coder™).