Project Machine learning

Detect butterfly

1. Dataset
   In this project, I train 3 kind of butterfly

   • CommonBuckeye
   • MourningCloak
   • ParisPeacock

   Total about 1200 images
   Butterfly images was taken from google and many pages.
   I created 3 utils for getting, converting type image, changing name.

   Script getting image from google image search

var cont=document.getElementsByTagName("body")[0];
var imgs=document.getElementsByTagName("a");
var i=0;var divv= document.createElement("div");
var aray=new Array();var j=-1;
while(++i<imgs.length){
  if(imgs[i].href.indexOf("/imgres?imgurl=http")>0){
    divv.appendChild(document.createElement("br"));
    aray[++j]=decodeURIComponent(imgs[i].href).split(/=|%|&/)[1].split("?imgref")[0];
    divv.appendChild(document.createTextNode(aray[j]));
  }
}
cont.insertBefore(divv,cont.childNodes[0]);

Script getting image from any page

function getAllImgageSrc() {
  let count = 0;
  let imgs = "";
  document.querySelectorAll("img").forEach(img=>{
    imgs+=img.src+'\n'
    count++;
  })
  console.log("Image length: ", count);
  console.log(imgs);
}

I used tool Yolo-Annotation-Tool-New to create the bounding box for images

1. Training
   • I followed to this link: @manivannan_data/how-to-train-yolov3-to-detect-custom-objects-ccbcafeb13d2"">https://medium.com/@manivannan_data/how-to-train-yolov3-to-detect-custom-objects-ccbcafeb13d2
   • I trained model on Google Colab follow this link: https://colab.research.google.com/drive/1lTGZsfMaGUpBG4inDIQwIJVW476ibXk#scrollTo=wkzMqLZV-rF5_
   • I did almost the same, fix some error about the path to file

Thank you for sharing <3

1. Notes
   Some points I got stuck, I’m not sure with others
   • Because I used google colab and got free 12GB GPU to training, So before make command after clone yolov3 project. I changed Makefile file: GPU = 1 and CUDNN=1
   • In butterfly-obj.data file. It will save a backup file and weight after 1000 at fist and 10000 times next by default
   • Because I don’t want save after 10000 times, so I change detector.c in examples folder. *Line 138 change to if(i%500==0 || (i < 1000 && i%100 == 0)){ ...}. It means it will save weights file after 100 times at first and 500 times for second time.
   • I saved backup file and wieghts in backup folder of google colab. So I need to download files before google reset

Finally, this is my notebok: https://colab.research.google.com/drive/1HAc0G3hedpFuJ1XI1fNVZSAYJGiaoCgq?authuser=1#scrollTo=wrI19JfCeyZh

Update

• Understand output: https://timebutt.github.io/static/understanding-yolov2-training-output/

Have a nice day!!!