Yolo-implementation-using-Intel-Movidius-Neural-Compute-Stick-NCS-on-Raspberry-Pi

Intel Movidius Neural Compute Stick (NCS) Setup

• Install NCSDK for the Movidius stick to work on Raspberry Pi.

  if it fails at :

  • If NCSDK fails to setup the Intel Movidius NCS download caffe from git if the installation fails

  • NCS also required ssd-caffe to run the model

          NOTE :
    

    • If any of the above mentioned or any other package suffers installation all the required file are available at movidius/*. All the files were tested.

News

• Works on both Rpi and DSI ( Rpi Camera Module)
• YOLOv1 Tiny is working.

Protobuf Model files

./prototxt/

Download Pretrained Caffe Models to ./weights/

• YOLO_tiny: https://pjreddie.com/media/files/yolov2-tiny.weights

Compilation

• Compile .prototxt and corresponding .caffemodel (with the same name) to get NCS graph file. For example: "mvNCCompile prototxt/yolo_tiny_deploy.prototxt -w weights/yolo_tiny_deploy.caffemodel -s 12"
• The compiled binary file "graph" has to be in main folder after this step.

Single Image Script

• Run "yolo_image_predict.py" to process a single image. For example: "python3 function/yolo_image_predict.py images/dog.jpg" to get detections as below.

Camera Input Script

• Run "yolo_video.py" to process a videos from your camera. For example: "python3 function/yolo_video.py" to get camera detections as below.
• Modify script arguments if needed.
• Press "q" to exit app.

$ NOTE : To run with external USB camera interfaced with Raspberry Pi, comment line #164 on function/yolo_video.py