aqsc / golov2

YOLOv2 for Golang

GOLOv2

YOLOv2 for Golang

This projects implements the yolov2 (https://pjreddie.com/darknet/yolov2/) RegionLayer in Go. It is heavily inspired by duangenquan's C++-RegionLayer implementation (https://github.com/duangenquan/YoloV2NCS).

This projects makes use of gocv (https://gocv.io) and go-ncs (https://github.com/hybridgroup/go-ncs/), both from hybridgroup (https://github.com/hybridgroup).

It comes with a tiny-yolo caffe model which I derived from original weights (https://pjreddie.com/media/files/yolov2-tiny-voc.weights) with this darknet2caffe converter: https://github.com/marvis/pytorch-caffe-darknet-convert. It also comes with a Movidius NCS model version of tiny-yolo which I compiled from the converted caffe model.

Setup

1. Install gocv as described on https://gocv.io/getting-started/
2. Install go-ncs as described on https://github.com/hybridgroup/go-ncs
3. Plug in your Movidius Neural Compute Stick
4. $ git clone [email protected]:dymat/GOLOv2.git
5. $ cd GOLOv2
6. go run *.go