training the network on ground-truth

[lasandcris]

I'd like to train the network using depth maps as ground truths. is that somehow possible - and if yes, can someone give me some pointers? thanks, laszlo

[piiswrong]

It's possible. remove the selection layer and quantize your depth map.

[lasandcris]

@piiswrong
could you please run through how to use the train.py?
thanks

[zedomel]

How to use train.py? I'm trying to create a database using parse.py and data.py but I got this error when run:
python parse.py movie.avi exp

Traceback (most recent call last): File "parse.py", line 204, in <module> process_movie2d(args.path, args.prefix) File "parse.py", line 184, in process_movie2d rect = list(get_clip_rect(fname, True)) File "parse.py", line 141, in get_clip_rect assert cap.read()[0] AssertionError

What should be the movie specs? For run parse.py and then train.py?

[lwkobe]

@zedomel Have you solved your problem? If you use the anaconda, you will find the conda-opencv3 which can't read a video. I am solving this problem.