CVPPP training

[arnike]

Hi Mengye,

thanks a lot for sharing the code!
I have a couple of questions about training on CVPPP:

1. In the paper you mention training on 128 images, but in cvppp.py I see that you also create train/val split (103/25). Did you use all 128 images or only 103 of them for test evaluation?
2. Are the pre-set numbers of iterations (60K for box net, 30K for full net) the ones used for test evaluation?
3. I noticed that the pre-processing net (FG+Angles) is not trained for the CVPPP in this implementation. Are the numbers reported in the paper for CVPPP with or without pre-processing?

Thanks in advance,
Nikita