calibrate-12 can not converge when trainning, while net-12 converged very easily about cnn_face_detection

[hiyijian]

besides, I noticed that you discussed the issue in issue#4.
It seems that the performance would be better if avoiding to use crop_size and setting mirror to false.
Maybe add more training patches?

[anson0910]

Yes, adding more training patches should help, I used about 500000 positive training samples.
Another issue is how good the initialization is, which is pure luck.
If it does not converge in 50000 iterations, it is likely that it will never converge.
Sometimes it just works by restarting the training!

[hiyijian]

Thanks. I used 1000000 patches, no luck so far. I will keep trying