Is it possible to generate Cuda code with the GPU Coder from a faster RCNN object Detector?

Hello,

I trained a faster RCNN detector on MATLAB to detect a custom object. This worked in MATLAB very well. As a next step I want to run the detector on a NVIDIA Jetson TX2. So I tried to generate CUDA code with the GPU Coder. But the Coder doesn't recognize the detect function. Is there a way to generate CUDA for object detection with a faster RCNN detector with the GPU Coder?

Thank you very much!

Answer 1

[Stefanie Schwarz]

As of MATLAB R2018b, we unfortunately do not support the R-CNN detection with GPU Coder. Currently, we only support the function 'matchFeatures' from the Computer Vision System Toolbox, but not 'detect'. We are moving forward to integrate this capability in future releases.

As for alternatives, we think YOLO will be your best bet. Alexnet and Squeeze will identify objects on the screen, but it does not provide the coordinates of the object.

Potentially you can try something like the Traffic Sign Detection and Recognition example where a YOLO-like DNN is used to detect objects, and a 2nd DNN is used to identify the object.

You may refer to the links below for other deep learning/object detection features currently supported by GPU Coder:

https://www.mathworks.com/help/gpucoder/gpucoder-deep-learning.html https://www.mathworks.com/help/gpucoder/examples/object-detection.html

[Antoine Lasfargeas]

Hello,

Is there nowadays any way to use the GPU coder with Faster R-CNN? I just want to deploy Faster R-CNN to an Nvidia Jetson Nano device.

Thank you in advance,