TrainImagesClassifier - Training Sample Size

I just have a question regarding the size of training samples used to train the image classifier. According to literature I need to take a small sample size of each class, but in this video Machine Learning For Remote Sensing : Orfeo ToolBox Meets OpenCV the training samples are very big.

Does the OpenCV library used by Orfeo classification have different requirement for training sample input? The more training samples, the better? What is the recommended training sample size for Orfeo Image Classifier?


The best training sample size depends on the algorithm used and on the input data. Maybe you can test several size, and compare the results (for example by also providing a validation sample).