How to prepare training data for MATLAB inbuilt multiclass Faster R-CNN function?

Question

I am training Faster-RCNN for multiclass object detection. I am using matlab inbuilt function trainFasterRCNNObjectDetector along with alexnet. The code works fine for single object but when tried for multiple objects it shows "Warning: Invalid bounding boxes from 2 out of 3 training images were removed. The following rows in trainingData have invalid bounding box data:". Its because some images doesn't contain one class and for that i just put empty brackets.

%trainingData
%|---------------------|------------------|------------------|
%|      Filename       |     Class 1      |     Class 2      |         
%|---------------------|------------------|------------------|
%|    '\00013.jpg'     |[1, 141, 374, 158]|        []        |
%|---------------------|------------------|------------------|
%
options = trainingOptions('sgdm', ...
    'MaxEpochs', 5, ...
    'MiniBatchSize', 1, ...
    'InitialLearnRate', 1e-3, ...
    'CheckpointPath', tempdir);

[detector, info] = trainFasterRCNNObjectDetector(trainingData, 'alexnet', options, ...
        'NegativeOverlapRange', [0 0.3], ...
        'PositiveOverlapRange', [0.6 1]);

Answer 1

Does this page help?

Based on the answer there, I think you have to put all of the coordinates for the objects of an image as a single entry in your training data table and separate them with a semicolon.