Using a trained FastRCNNObjectDetector, expected LABEL to be nonempty

2 次查看（过去 30 天）
SALEM ALMARRI
SALEM ALMARRI 2019-3-9
评论： Jingang Wu 2019-11-10
Hello,
I used the following trained Fast R-CNN:
trainingData = objectDetectorTrainingData(gTruth)
net = vgg16;
layersTransfer = net.Layers(1:end-3);
% numClasses = numel(categories(imdsTrain.Labels));
layers = [
layersTransfer
fullyConnectedLayer(9,'WeightLearnRateFactor',10,'BiasLearnRateFactor',10)
softmaxLayer
classificationLayer];
options = trainingOptions('sgdm', ...
'MiniBatchSize', 1, ...
'InitialLearnRate', 1e-6, ...
'MaxEpochs', 10)
frcnn = trainFastRCNNObjectDetector(trainingData, layers , options, ...
'PositiveOverlapRange',[0.1 1], ...
'NegativeOverlapRange',[0 0.1])
The training was based on trainingData which was a an imported gTruth table which included around 8 different ROI labels for 4,000 images of 224x224 dimension.
Then I have saved the trained FRCNN to classify a sample image, the error I have getting is "Expected LABEL to be nonempty", and in workspace the labels is indeed empty, I am confused as to why this has happened though I have trained the network and everything.
load frcnn
detector = frcnn
I = imread('sampleImage.jpg')
imshow(I)
[bboxes,scores,labels] = detect(detector,I)
detectedI = insertObjectAnnotation(I,'Rectangle',bboxes,cellstr(labels))
figure
imshow(detectedI)
  1 个评论
Jingang Wu
Jingang Wu 2019-11-10
I used the pretrained detector in mathwork example "Object Detection Using Faster R-CNN Deep Learning". I encountered the same problem. I applied my own dataset and for the first image it was working well. But for other images, the bboxes was empty.

请先登录，再进行评论。

请先登录，再回答此问题。

回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Object Detection 的更多信息

标签

产品


版本

R2018b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by