add 3 models

[dereyly]

1x5_all -- in model "5x5_to_3x3_to_1x3_AND_faster_lr_AND_PReLU" replace all 1x3 to 1x5. Complex same as 3x3+1x1

1x3_sz_augm -- model "5x5_to_3x3_to_1x3_AND_faster_lr_AND_PReLU" data layer use size augmentation. There are a few changes in caffe-elu.

darknet_bn -- model darknet from fast YOLO https://github.com/pjreddie/darknet/blob/master/cfg/darknet.cfg. Extend LSUV init, PreLU and BN

[ducha-aiki]

Thanks :)

1x3_sz_augm starts only from 60K iterations...Is the start available for it?

[dereyly]

before 60k iteations i have failed logs with test data. I use standard dataset for test without scaling 0.04 and have 0% accuracy

[ducha-aiki]

OK

[ducha-aiki]

DarkNet is interesting. Thanks!