Question about YOLO implementation in MATLAB

Hi,

Need some help from any experienced developers regarding YOLO in MATLAB.

I am trying to build the YOLO (You Only Look Once) neural network on MATLAB for an autonomous driving project. Some parts of the project has already been implemented based on this guide:

https://www.mathworks.com/help/gpucoder/examples/object-detection.html?requestedDomain=www.mathworks.com

The example shown imports YOLO from Caffe, then proceeds to convert it to CUDA implementation. I have downloaded and converted the YOLO from Caffe to MATLAB, and successfully obtained the output using MATLAB's predict() function. However, I am stuck as i cannot understand how to decipher the 1x1470 vector produced by YOLO and use it for object detection. I understand that the 1x1470 vector corresponds to the 7x7x30 vector produced by the original YOLO and have managed to produce the output, but it does not seem to work as the outputs were all close to 0, which does not make sense. So, my questions are:

1. How do you decipher the MATLAB YOLO's 1x1470 vector into meaningful object detections (bounding boxes, confidence, etc.)?

2. Why are all the outputs close to zero? From the example it mentioned that I am downloading a pretrained model, so there should be no issues with this.

Thanks in advance.

IMY

[Watcharin Tangsuksant]

What MATLAB version did you use for YOLO?

[Ihsan Yassin]

R2017b windows x64. Any ideas?

[Watcharin Tangsuksant]

Sorry, I don't have idea becuase I just study about YOLO in Matlab and I don't know how to use it.

So, I have some question .... Can Matlab2016a run the YOLO? and do you have the instruction or URL for installing the YOLO in Matlab?

Thank you.

[Watcharin Tangsuksant]

What file did you import and convert? As you mentioned that " imports YOLO from Caffe, then proceeds to convert it to CUDA implementation", Can you tell me about that file or URL?

Answer 1

[pony candy]

the outcome is a tensor you gonna need yetensor toolbox to deal with that