Skip to content

Latest commit

 

History

History
141 lines (105 loc) · 3.22 KB

ReadMe.md

File metadata and controls

141 lines (105 loc) · 3.22 KB
Raw

Overview

This is the python script, which is server of detection using darknet yolo.

Darknet is as following:

This server has two API.

  1. Getting the detection result as json format including image(base64 decoded) URI: '/detect'

    parameter example

    {
  "image": base64.b64decode(imgdata).decode("utf-8"),
  "thesh": 0.5,
  "get_img_flg": True

    The parameter of 'thresh' is threshold of yolo. This is not required and the default value is 0.5. If thresh is less than default, more object can be detected but misrecgnition increses.

    The paramete of get_img_flg is boolean. If true, return image data embeded bbox of yolo resuls. This parameter is optional and the defalut value is true.

SoftWare requirements

  • Python > 3.6

    I test only 3.7.1

  • darknet revision: 9a4b19c4158b064a164e34a83ec8a16401580850 darknet revision: 61c9d02ec461e30d55762ec7669d6a1d3c356fb2

Setup

Install Directly Your Machine

  1. Install darknet

    Install darknet, reading the website of darknet

  2. Clone of this repository

    $ git clone https://github.com/komorin0521/darknet_server

  3. Move all files and folders into darknet folder

    $ cp -r darknet_server/* ${darknetPATH}/python

  4. Install python module using pip

    $ (sudo) pip install -r requirements.txt

  5. Running server

    $ python3 darknet_server.py

If you want to your original model, please modify 'darknet_server/conf/darknet_server.ini'

Using docker

requirements

  1. docker-compose version 1.19.0-rc3, build cdd0282
  2. Docker version 18.06.1-ce, build e68fc7a. installed nvidia-docker

I use the cuda-9.2 and nvidia driver is 396.44 If you use another version, please modify the source image of dokcer, "nvidia/cuda:9.2-devel-ubuntu16.04"

How to install

  1. install docker, nvidia-docker, docker-compose

  2. please check the run of nvidia-smi

  3. start docker

    $ docker-compose up

Check the server response from other terminal

$ python simple_client.py |sed "s/'/\"/g" |jq

If the server work well, you will get message like following

{
  "status": "success",
  "resultlist": [
    {
      "class_index": 0,
      "obj_name": "person",
      "score": 0.9997863173484802,
      "bounding_box": {
        "x_min": 191.12104034423828,
        "y_min": 100.11891174316406,
        "width": 79.28132629394531,
        "height": 277.3841247558594
      }
    },
    {
      "class_index": 17,
      "obj_name": "horse",
      "score": 0.9993027448654175,
      "bounding_box": {
        "x_min": 402.7898635864258,
        "y_min": 135.62349700927734,
        "width": 198.86619567871094,
        "height": 219.7958526611328
      }
    },
    {
      "class_index": 16,
      "obj_name": "dog",
      "score": 0.9974156618118286,
      "bounding_box": {
        "x_min": 60.97343444824219,
        "y_min": 264.89939880371094,
        "width": 138.28082275390625,
        "height": 82.55728149414062
      }
    }
  ]
}

More information

If you want more information about scripts, please show the help as following.

$ python server.py -h

Contacts

If you have any questions, please send me an email or pull request.

email: [email protected]