How can I use the output from Graph.GetResult() in order to show image with all detections boxes?

[idata]

As you can see in the attached screen shot the output from this function (GetResuls) is an array with length 5070 . Please if someone can help me understand what is the meaning of this output and how can I use those points to show the detection bounding boxes. thanks for all :)

[idata]

@veredpoz, I am guessing you are working with either TinyYOLO or MobileNet SSD. The output array format depends on which of the two networks you are using. You can see a description of the format here:

MobileNet SSD: https://github.com/movidius/ncappzoo/blob/master/caffe/SSD_MobileNet/run.py#L50

TinyYOLO: https://github.com/movidius/ncappzoo/blob/master/caffe/TinyYolo/run.py#L22

If you are working on MobileNet SSD, I have created a utility wrapper that formats the output array into a Python dictionary. Try it out and let me know if it helps.

Sample app: https://github.com/ashwinvijayakumar/ncappzoo/tree/detector-template/apps/ssd-object-detector

Utility wrapper: https://github.com/ashwinvijayakumar/ncappzoo/blob/detector-template/apps/ssd-object-detector/utils/deserialize_output.py

[idata]

@AshwinVijayakumar Thanks Ashwin,

1. I am using Tiny-Yolo V2 through darkflow transform written in tensorflow. I trained it to detect a single object class. How to change the example given in TinyYOLO: https://github.com/movidius/ncappzoo/blob/master/caffe/TinyYolo/run.py#L22 to work with a single class?
2. Is it possible to compile Mobilenet-SSD written in tensorflow?

[idata]

@veredpoz For our version of Tiny Yolo, the inference result can be interpreted as in my post https://ncsforum.movidius.com/discussion/comment/2172/#Comment_2172.

[idata]

@Tome_at_Intel

1. Thank you for your response. my problam is that I get much more outputs, so that output[1078:] contains 4802 parameters and it can't be converted the shape of (7,7,2,4).
2. How can I use the code you have attached for one class only? I get output of 5880 length.

[idata]

@veredpoz Where did you get this model from? The size of the output can vary depending on how many anchor/prior boxes you specify and how many grid cells you have. I believe ours is a 7x7 grid with 2 bounding boxes per grid cell.

[idata]

Yes, I took the same model, but I trained it on my own dataset and I only have one class so I changed the amount of filters in the final layer to 30