Someone asked me before. This is actually a very simple question. You can generate a simple cascade_data script.
But later I found that if I did not read it carefullyCodeAnd is very prone to problems. Here is a brief introduction to help new students.
First, the positive sample used to generate PCA statistics (no matter what the negative sample is) must be consistent with the positive sample used in the training model.
That is to say, the sample path and quantity in pascal_data cannot be changed! Pascal_data is called in cascade_data.
Steps:
1 open MATLAB and enter VOC-release, addpath star-cascade
2 load ('your _ model ')
3 cascade_data (model, 'Year', 0) # If the sample is prepared by yourself, the string corresponding to year and model. Year is the same, and the third parameter must be 0!
4 cascade_data (model, 'Year', 5) # The number of PCA instances is random.
You can run cascade_demo.
I have to say that the Framework Code is too messy. However, my c ++ detection code is OK.