Main Content

evaluateDetectionPrecision

(To be removed) Evaluate precision metric for object detection

Use evaluateObjectDetection instead of evaluateDetectionPrecision, which will be removed in a future release. The more recent evaluateObjectDetection can be used to perform a comprehensive analysis of object detector performance.

Description

averagePrecision = evaluateDetectionPrecision(detectionResults,groundTruthData) returns the average precision, of the detectionResults compared to the groundTruthData. You can use the average precision to measure the performance of an object detector. For a multiclass detector, the function returns averagePrecision as a vector of scores for each object class in the order specified by groundTruthData.

example

[averagePrecision,recall,precision] = evaluateDetectionPrecision(___) returns data points for plotting the precision–recall curve, using input arguments from the previous syntax.

[___] = evaluateDetectionPrecision(___,threshold) specifies the overlap threshold for assigning a detection to a ground truth box.

Examples

collapse all

This example shows how to evaluate a pretrained YOLO v2 object detector.

Load the Vehicle Ground Truth Data

Load a table containing the vehicle training data. The first column contains the training images, the remaining columns contain the labeled bounding boxes.

data = load('vehicleTrainingData.mat');
trainingData = data.vehicleTrainingData;

Add fullpath to the local vehicle data folder.

dataDir = fullfile(toolboxdir('vision'), 'visiondata');
trainingData.imageFilename = fullfile(dataDir, trainingData.imageFilename);

Create an imageDatastore using the files from the table.

imds = imageDatastore(trainingData.imageFilename);

Create a boxLabelDatastore using the label columns from the table.

blds = boxLabelDatastore(trainingData(:,2:end));

Load YOLOv2 Detector for Detection

Load the detector containing the layerGraph for trainining.

vehicleDetector = load('yolov2VehicleDetector.mat');
detector = vehicleDetector.detector;

Evaluate and Plot the Results

Run the detector with imageDatastore.

results = detect(detector, imds);

Evaluate the results against the ground truth data.

[ap, recall, precision] = evaluateDetectionPrecision(results, blds);

Plot the precision/recall curve.

figure;
plot(recall, precision);
grid on
title(sprintf('Average precision = %.1f', ap))

Train an ACF-based detector using preloaded ground truth information. Run the detector on the training images. Evaluate the detector and display the precision-recall curve.

Load the ground truth table.

load('stopSignsAndCars.mat')
stopSigns = stopSignsAndCars(:,1:2);
stopSigns.imageFilename = fullfile(toolboxdir('vision'),'visiondata', ...
    stopSigns.imageFilename);

Train an ACF-based detector.

detector = trainACFObjectDetector(stopSigns,'NegativeSamplesFactor',2);
ACF Object Detector Training
The training will take 4 stages. The model size is 34x31.
Sample positive examples(~100% Completed)
Compute approximation coefficients...Completed.
Compute aggregated channel features...Completed.
--------------------------------------------
Stage 1:
Sample negative examples(~100% Completed)
Compute aggregated channel features...Completed.
Train classifier with 42 positive examples and 84 negative examples...Completed.
The trained classifier has 19 weak learners.
--------------------------------------------
Stage 2:
Sample negative examples(~100% Completed)
Found 84 new negative examples for training.
Compute aggregated channel features...Completed.
Train classifier with 42 positive examples and 84 negative examples...Completed.
The trained classifier has 20 weak learners.
--------------------------------------------
Stage 3:
Sample negative examples(~100% Completed)
Found 84 new negative examples for training.
Compute aggregated channel features...Completed.
Train classifier with 42 positive examples and 84 negative examples...Completed.
The trained classifier has 54 weak learners.
--------------------------------------------
Stage 4:
Sample negative examples(~100% Completed)
Found 84 new negative examples for training.
Compute aggregated channel features...Completed.
Train classifier with 42 positive examples and 84 negative examples...Completed.
The trained classifier has 61 weak learners.
--------------------------------------------
ACF object detector training is completed. Elapsed time is 29.6068 seconds.

Create a table to store the results.

numImages = height(stopSigns);
results = table('Size',[numImages 2],...
       'VariableTypes',{'cell','cell'},...
       'VariableNames',{'Boxes','Scores'}); 

Run the detector on the training images. Store the results as a table.

for i = 1 : numImages
    I = imread(stopSigns.imageFilename{i});
    [bboxes, scores] = detect(detector,I);
    results.Boxes{i} = bboxes;
    results.Scores{i} = scores;
end 

Evaluate the results against the ground truth data. Get the precision statistics.

[ap,recall,precision] = evaluateDetectionPrecision(results,stopSigns(:,2));

Plot the precision-recall curve.

figure
plot(recall,precision)
grid on
title(sprintf('Average Precision = %.1f',ap))

Input Arguments

collapse all

Object locations and scores, specified as a two-column table containing the bounding boxes and scores for each detected object. For multiclass detection, a third column contains the predicted label for each detection. The bounding boxes must be stored in an M-by-4 cell array. The scores must be stored in an M-by-1 cell array, and the labels must be stored as a categorical vector.

When detecting objects, you can create the detection results table by using imageDatastore.

        ds = imageDatastore(stopSigns.imageFilename);
        detectionResults = detect(detector,ds);

Data Types: table

Labeled ground truth, specified as a datastore or a table.

Each bounding box must be in the format [x y width height].

  • Datastore — A datastore whose read and readall functions return a cell array or a table with at least two columns of bounding box and labels cell vectors. The bounding boxes must be in a cell array of M-by-4 matrices in the format [x,y,width,height]. The datastore's read and readall functions must return one of the formats:

    • {boxes,labels} — The boxLabelDatastore creates this type of datastore.

    • {images,boxes,labels} — A combined datastore. For example, using combine(imds,blds).

    See boxLabelDatastore.

  • Table — One or more columns. All columns contain bounding boxes. Each column must be a cell vector that contains M-by-4 matrices that represent a single object class, such as stopSign, carRear, or carFront. The columns contain 4-element double arrays of M bounding boxes in the format [x,y,width,height]. The format specifies the upper-left corner location and size of the bounding box in the corresponding image.

Overlap threshold for assigned a detection to a ground truth box, specified as a numeric scalar. The overlap ratio is computed as the intersection over union.

Output Arguments

collapse all

Average precision over all the detection results, returned as a numeric scalar or vector. Precision is a ratio of true positive instances to all positive instances of objects in the detector, based on the ground truth. For a multiclass detector, the average precision is a vector of average precision scores for each object class.

Recall values from each detection, returned as an M-by-1 vector of numeric scalars or as a cell array. The length of M equals 1 + the number of detections assigned to a class. For example, if your detection results contain 4 detections with class label 'car', then recall contains 5 elements. The first value of recall is always 0.

Recall is a ratio of true positive instances to the sum of true positives and false negatives in the detector, based on the ground truth. For a multiclass detector, recall and precision are cell arrays, where each cell contains the data points for each object class.

Precision values from each detection, returned as an M-by-1 vector of numeric scalars or as a cell array. The length of M equals 1 + the number of detections assigned to a class. For example, if your detection results contain 4 detections with class label 'car', then precision contains 5 elements. The first value of precision is always 1.

Precision is a ratio of true positive instances to all positive instances of objects in the detector, based on the ground truth. For a multi-class detector, recall and precision are cell arrays, where each cell contains the data points for each object class.

Version History

Introduced in R2017a

expand all