Hi Leonard,
The error occurs because bagOfFeatures is extracting features from all of your training images and storing them in memory. This is a simple strategy for collecting the features, but does have the downside of requiring lots of memory.
If you're using the 'Grid' point selection method, then you can try increasing the GridStep parameter [32 32], which will reduce the number of locations within the image where features are extracted. In addition to that, you can try reducing the number of blocks using the 'BlockWidth' parameter, for example, set it to [64 96].
Another common strategy to avoid this type of problem with large datasets is to sub-sample the number of features extracted from the training images.
One way to do this is to define a custom extractor function that selects only a fraction of the features from an image. There is an example custom feature extractor that you can look here:
You can see the M-code by editing the example: edit('exampleBagOfFeaturesExtractor.m')
This example uses the same feature extraction method that the bagOfFeatures uses by default.
If you look on line 69, you'll see where the features are extracted. Then on line 79, the strength of the features is computed.
Using the strength information, you can select just the top 50% of the features from the image and return those instead of returning all of them. For example, add the following code after line 79:
% sort features by feature metric
[featureMetrics, idx] = sort(featureMetrics, 'descend');
features = features(idx, :);
% keep only the top 50%
numFeatures = size(features,1);
numFeaturesToKeep = floor(0.5 * numFeatures);
features = features(1:numFeaturesToKeep, :);
If none of these work, then try reducing the number of training images.
Hope that helps, Birju
