/Pill

Pill project

Pill

This project is not intented for public use so all the environemnt setting and installation procedure will not work in other computer

Requirement

  • The following components are already installed in Mindspree deep learning server
  • Darknet Yolo
  • Linux PC with Nvida GPU
  • OpenCV for Linux (Optional)

How to train

In order to train darkent for this application, you need to follow the steps below.
Please go to the reference site and read it through. This is very important
( Reference site: https://timebutt.github.io/static/how-to-train-yolov2-to-detect-custom-objects/ )

  • Prepare for the training and testing data

  • Training image file path
    Place all the images and annotation file in this folder. Each image file is paried with annotation file. Please check the image below ~/Github/darknet/data/img

    The annotation field structure is as follows

    • [category number] [object center in X] [object center in Y] [object width in X] [object width in Y]
      In our case, we only detect pill so there is only one category. The category will be alwyas 0

    • For easier annotation, you need to use the following tool
      https://github.com/Lab930boss/Yolo_mark

Alt text

  • Set up obj.data file
    In this file, training and validation file paths are defined.
    ~/Github/darknet/data/obj.data

    Contents

    • classes =1
    • train = data/train.txt
    • valid = data/train.txt --> In this case train data is used for validation
    • names = data/obj.names
    • backup backup

  • Set up cfg/yolo-obj.cfg that defines DNN structure

  • Training command ~/Github/darknet/darknet detector train data/obj.data cfg/yolo-obj.cfg darknet19_448.conv.23

    • If darknet doesn't work, use darknet_train instead
    • The darknet19_448.conv.23 is the pretrained network

    Note: When you edit cfg file please be careful about the folliwing rule set classes=1, the number of categories we want to detect set filters=(classes + 5)*5 in our case filters=30

  • Test command ~/Github/darknet/darknet detector test data/obj.data cfg/yolo-obj.cfg yolo-obj1000.weights data/testimage.jpg
    - If darknet doesn't work, use darknet_train

** Note Due to compiing issue, please use ~/Github/darknet2018/darknet instead of ~/Github/darknet