Tool for generating eye-negative face samples for cascading. This code uses the default pre-trained OpenCV models in order to generate eye-negative face samples for use in further eye detection model cascade training. It is fairly cluttered and not safe for production but serves as a good inspiration for others that may need something similar. A number of constraints such as the difference between eye sizes and eye alignment have been set in order to ensure that the generated samples are 99% valid. Manual verification on thousands of samples have yielded these results, but don't take my word for it.
These negative samples have been generated using the Selfie Dataset provided by Kalayeh, Mahdi M and Seifu, Misrak and LaLanne, Wesna and Shah, Mubarak.
- cleanup and modularity via command line options
- command line option for face crop or bw generation