Object Detection Using YOLO v2 Deep Learning

I want to train a YOLO v2 Object Detection Network for pictures size of 4000*3000 where a have labeled very small objects,what would be the optimal size for the inputSize? As if i use the original image size(4000*3000) my computer crashes,but if i use the specified size(224*224) my Yolo2ObjectDetector does not detect any objects on a test image.

inputSize = [224 224 3];
numClasses = width(vehicleDataset)-1;
trainingDataForEstimation = transform(trainingData,@(data)preprocessData(data,inputSize));
numAnchors = 7;
[anchorBoxes, meanIoU] = estimateAnchorBoxes(trainingDataForEstimation, numAnchors);
featureExtractionNetwork = resnet50;
featureLayer = 'activation_40_relu';
lgraph = yolov2Layers(inputSize,numClasses,anchorBoxes,featureExtractionNetwork,featureLayer);

options = trainingOptions('sgdm',...
          'InitialLearnRate',0.001,...
          'Verbose',true,...
          'MiniBatchSize',16,...
          'MaxEpochs',30,...
          'Shuffle','never',...
          'VerboseFrequency',30,...
          'CheckpointPath',tempdir);
[detector,info] = trainYOLOv2ObjectDetector(trainingDataForEstimation,lgraph,options);

Answer 1

[Shishir Singhal]

Hi,

YOLO v2 object detector has a problem with detecting small objects. Instead you can try YOLO v3. It is far good for detecting small objects. You can refer to this documentation: https://www.mathworks.com/help/vision/examples/object-detection-using-yolo-v3-deep-learning.html to know more about its implementation in MATLAB.

Hope this helps....!!!