Threshold for evaluation the R-CNN detector

Question

Abdussalam Elhanashi 2019-9-19

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/481177-threshold-for-evaluation-the-r-cnn-detector

评论： Abdussalam Elhanashi 2019-9-23

Hi Guys

I would like if possible how to make this Treshold for Evaluation and validation of created R-CNN object Detector, i tried to make it in the attached scripts but it does not work, I want to make Threshold for score that like below 0.58 that score and bboxes should not be appeared

Herein the code:-

load('gTruth.mat')
smokedetection = selectLabels(gTruth,'smokealarm');
if ~isfolder(fullfile('EvaluationData'))
    
    mkdir EvaluationData
    addpath('EvaluationData');
    
    evaluationData = objectDetectorTrainingData(gTruth,...
    'SamplingFactor',1,'WriteLocation','EvaluationData');
end
imds = imageDatastore(fullfile('EvaluationData'));
numImages = height(evaluationData);
result(numImages,:) = struct('Boxes',[],'Scores',[]);
for i = 1:numImages
    
    % Read Image
    I = readimage(imds,i); 
    
    % Detect the object of interest
    [bboxes, scores] = detect(detector,I,'Threshold',1);
    
    % Store result 
    result(i).Boxes = bboxes;
    result(i).Scores = scores;
   
end
% Convert structure to table
results = struct2table(result);
overlap = 0.1;
% Evaluate Metrics
[ap,recall,precision] = evaluateDetectionPrecision(results...
    ,evaluationData(:,2),overlap);
[am,fppi,missRate] = evaluateDetectionMissRate(results,evaluationData(:,2),overlap);
% Plot Metrics
subplot(1,2,1);
plot(recall,precision);
xlabel('Recall');
ylabel('Precision');
title(sprintf('Average Precision = %.1f', ap))
grid on
subplot(1,2,2);
loglog(fppi, missRate);
xlabel('False Positives Per Image');
ylabel('Log Average Miss Rate');
title(sprintf('Log Average Miss Rate = %.1f', am))
grid on

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Harsha Priya Daggubati 2019-9-23

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/481177-threshold-for-evaluation-the-r-cnn-detector#answer_393030

Hi,

One possible way to get only the scores returned by ‘detect’ greater than a value, say 0.58 is to store only the score value greater than 0.58 and its corresponding bboxes in result and use it in your evaluation metrics. You can also try using Overlap Threshold option for ‘evaluateDetectionPrecision’function as mentioned in the following documentation link.

https://www.mathworks.com/help/vision/ref/evaluatedetectionprecision.html?s_tid=doc_ta