I'm going to participate in a team competition in robotics, which involves identifying and interacting with a number of objects in the surrounding, amongst others a red button and a human.
To detect those objects, we are going to use a camera, and a deep convolutional neural network to interpret the camera image. For the main computational device, we are probably going to use an Android phone, so the deep learning framework has to run on Android.
Also, since we want the network to be as fast as possible, we want the framework to run on the GPU if possible. And since I guess most Android devices don't have NVIDIA graphics cards, being able to run on the GPU means using OpenCL (alternatively GLSL, but that seems unlikely).
To sum this up, I would like a deep learning framework that:
- Runs on Android devices, and
- Uses OpenCL (or GLSL) to do the heavy computations.
Is there any such deep learning framework?