Serverside inference for object annotation using the HoloLens 1. The HoloLens client is here.

This code is forked from Darknet by Joseph Redmon
https://github.com/pjreddie/darknet

Differences from Original

Changes made to original darknet code (at the time) are in src/demo.c.
server.h and server.c contain functionality to interface with a HoloLens client.

Installation

If you can, just use nvidia-docker.

If not, you need to first install OpenCV 2.4.13, CUDA 8, and Cudnn 5 (other versions may work, but I've only tested with these on linux).

Then compile with

make

For more details, check out the darknet installation guide or the original repo

To Run:

To run the server, use:

./darknet detector demo cfg/coco.data cfg/yolo.cfg yolo.weights

To run the server without having to hassle with old cuda and opencv versions (you really don't), with Nvidia Docker installed, use:

docker build -t yolo-server .
docker run --runtime=nvidia -p 11000:11000 -it yolo-server