Revision history [back]

Please take a look at the documentation and tutorial on training OpenCV models using cascade classifiers. Basically you need to choose to either, provide the images as a list of images, and then magically generate positives from that OR you decide to give it a file with the exact locations of objects and only retrieve those.

You file clearly indicates you want the second approach, telling the application that you want 1200 positives while you only have 28 image names there with literally 1 bounding box, so it continous and tries to read the following line, which is not there and just crashes.

So please, read again the tutorial. For now you can even use the updated version, which is still under reviewing!