如何计算物体检测/识别的混淆矩阵？答案

【问题标题】：How to calculate confusion matrix for object detection/recognition?如何计算物体检测/识别的混淆矩阵？
【发布时间】：2019-12-05 18:23:41
【问题描述】：

我已经解决了总共 32 个类别的一个图像识别问题。我得到了结果并计算了它的平均精度。我需要绘制混淆矩阵。

【问题讨论】：

标签： matlab plot image-recognition confusion-matrix

【解决方案1】：

我正在研究 YOLOv2，并为检测网络创建了一个混淆矩阵。希望这会有所帮助:)

testObjects 是真实标签，predLabels 是预测标签
TestData 是 imdsTest 的 imageDatastore()

testObjects = testData.UnderlyingDatastores{1, 1}.Files  ; %'C:\Users\admin\Desktop\Img_Data\Flower1\Flower101.jpg'
testObjects = erase(testObjects,fullfile(pwd,imgFolderName)); %'\Flower1\Flower101.jpg'
testObjects = categorical(extractBetween(testObjects, "\","\")); % Flower1 - array

predLabels = zeros(2,1);  
predLabels = categorical(predLabels); % Prelocation
for iPred = 1:length(testObjects)  
    [~, idxx] = max(cell2mat(detectionResults.Scores(iPred))); % max of all the bounding box scores
    multiLabels = detectionResults.Labels{iPred,1};  % find label of max score
    if isempty(multiLabels) == 1  
        predLabels(iPred,1) = {'NaN'};  
        predLabels(iPred,1) = standardizeMissing(predLabels(iPred,1),{'NaN'});  
    else  
       predLabels(iPred,1) = (multiLabels(idxx,1));  
    end  
end  
predLabels = removecats(predLabels);

plotconfusion (testObjects,predLabels) %confusionchart(testAsts,predLabels)

【讨论】：