Revision history [back]

first, please have a look at the tutorial

-info is actually an input argument, you're expected to give a file with image names and annotated positive rects there, it will not produce any output file.

also, you're lacking the mandatory -vec some_file argument where the desired positive vector should go, and values for -wand -h (the desired training window size).

then, trying to "synthesize" multiple positive samples from a single positive image usually does not produce any good results, so instead:

• collect many positive images (as many as you can)
• use the opencv_annotation tool (from apps folder) to generate a pos.txt with imagenames and annotations
• use that file with the -info argument with create_samples
• proceed with train_cascade

good luck !