Closed Verichev closed 5 years ago
I'm approving this one for check-in, but am working on creating a data collector node for actually writing the data. Meanwhile, you guys can continue with training.
@bessszilard Check the green path, if it slowes down, the car will not catch up with planned path and go off the road. Even without saving I have sometime this problem. This only work good on @safdark computer, cause he has raw linux. If you use this branch for saving images, use manual driving, it works perfectly. Than use images for feeding the model
You are right @Verichev . Probably the issue was with my machine. When the green path slowed down, I disabled the camera, and immediately the green path speed up, and continued everything as it have to be.
I am using Ubuntu 16.04 LTS on my notebook with kinetic 1.12.14.
This is probably because the io bottleneck is creating a race condition. As I had mentioned earlier, another node should be writing to disk, otherwise tl_detector has too much going on. It will also help with debugging if we separate the responsibilities of nodes this way.