Remember to normalize your data to -1,1]: outputs AND input images. Also ReLU layers are your friend.
(if you’re talking about using neural networks for vision)
I think neural networks are overkill for this challenge. It’s much easier and just as effective to run images through some filters in NI IMAQ or OpenCV and process the results without them.
Is anyone legitimately getting into deep learning?
If you are, I’d highly recommend reevaluating your priorities.
I think if you’ve got everything else covered already, deep learning isn’t a bad thing to learn about.
Just stick lots of pictures of your SPY in full uniform but in various lighting into TensorFlow. Then stick lots of pictures of the driver’s stations (including team #'s) into TensorFlow. Then massage out a probabilistic model of absolute position after DEFENSE crossing in autonomous.
Easy-peasy, right? Right?
Deep learning with things like neural networks is definitely something I’d recommend looking into. Just not for problems this simple.