(probably a) Noob-question regarding object-detection + CUDA

Hi,

I work for a non-profit association on a small since-project: the goal is to measure the speed of fishes in a pond or a tank filmed from above with a 720p IP cam. Major probs are different lighting conditions (eg. shadows, sunlight, reflections on the water-surface…).

I’ve not worked with commercial tools nor CUDA yet. My Python is okay’ish at best. Is there a way for a noob as me to get something running? :) As far as I understand it the most CPU-intensive work is the training of the neural network? My PC got a GTX970 and 16GB RAM, if that is enough? Time for the learning-phase isn’t the biggest prob. What about the actual detection? Does it still need overwhelming CPU-power or can I get some kind of AI-accelerater to get this running on a say… Pi3? Or atleast Bananapi or Intel NUC?

Happy new year and thank you in advance!

kind regards

Hi,

Happy New Year.

It’s recommended to check our tutorial first:
https://github.com/dusty-nv/jetson-inference

Although it is a C++ sample, you can get a better good solution by following the instructions step by step.
If training time is not important for you, your GPU should still be able to fulfill your requirement.

Thanks.