I have a problem with my detector , i get [bbox, score, label] empty.

%% detection

pp=alexnet;

pp1=pp.Layers;

pp=pp.Layers(1:19);

ppp=[pp

fullyConnectedLayer(2)

softmaxLayer()

classificationLayer()];

options = trainingOptions('sgdm', ...

'MiniBatchSize', 10, ...

'InitialLearnRate', 1e-3, ...

'MaxEpochs', 1, ...

'CheckpointPath', tempdir);

train1 =trainFastRCNNObjectDetector(gTruth, ppp, options, ...

'NegativeOverlapRange', [0 0.1], ...

'PositiveOverlapRange', [0.5 1], ...

'SmallestImageDimension', 300);

img = imread('image (825).JPG');

[bbox, score, label] = detect(train1, img);

imshow(insertObjectAnnotation(img, 'rectangle', bbox, label));

Answer 1

[Shuba Nandini]

Hello,  

 

It is my understanding that you want to train the “trainFastRCNNObjectDetector” with ‘alexnet’ as the backbone network. 

As per the documentation, “trainFastRCNNObjectDetector” function offers a functionality to automatically transform the backbone classification network, into a Fast R-CNN network by adding an ROI max pooling layer, classification layer and regression layer. 

The above functionality can be achieved, by specifying the required classification network name for the “network” argument. 

Please refer to the following link, for further information, 

https://www.mathworks.com/help/vision/ref/trainfastrcnnobjectdetector.html#mw_a57cdc64-ff65-4a22-b6ce-e3479faef270 

Hope this helps! 

Regards, 

Shuba Nandini