Yolo5 Training - Basics

This is the third page on the basic use of a detection model using

  • Our own images
  • Labeling our own images
  • Extracting the object information from each image file and creating .txt file for each image filename
  • Now that our data is ready with a training and testing sets we are ready to proceed training and testing with Yolo
  • First we need to setup a YAML file, we will name it data.yaml

Setup YAML file

The YAML file will contain all the information regarding the classes such as this

  • As you see we provide the model with the path to the training and testing data
  • We provide the name count as nc (20 in this case)
  • List of the names we have in the images

GPU

It’s much more expedient to run your models on a GPU, so if you don’t have access to one, Google Colab offers the use of one on a free basis.

Let’s pretend we will use Google Colab

Google Colab

  • From google drive > add more apps
  • Install colab
  • If you want to, access colab directly from: colab.research.google.com
  • Or from Drive > right click > new folder > name it: yolo_train > next
  • To create colab notebook: Open the file > from inside the folder, Right Click > Colab
  • Rename the file, it will have ipynb extension
  • Refresh the page and you’ll see the new name of the file

Connect to GPU

  • With the file open > Menu Tab: Runtime > Change Runtime Type
  • By default : None
  • To connect to GPU/TPU you have 12 hours which might not be applicable
  • Click on the option > save
  • In the upper right on the menu bar > Connect > It will change status and show the RAM and Disk resources
  • When you are not running anything > Terminate session
  • Idle time is counted as well

Terminate session: whenever you are not running anything on the GPU, Terminate session: From Menu bar > Runtime > Manage Sessions > TERMINATE

Connect Colab to Drive

  • This is pretty much already connected but in case
  • If you are inside colab and want to access notebooks in Drive
  • Left panel > Folder > Now look in Left Top bar > Folder > Mount Drive Option > Connect Drive appears when you scroll over folder > Click on it
  • Now Drive will appear in the File Tree > access the files in Drive
  • Remember we are working from inside the notebook

Extract Path to Notebook

  • In order to direct the model to our notebook/code we need to extract the path to our notebook
  • We can do that with by > expand the file tree to get to out notebook > right click > copy path > insert path in the () below
  • so what we have done is change directories to where our notebook is
  • We can verify that by using ls
import os
os.chdir('path to notebook in Drive')
# This will list the files in the current directory we just changed to, so let's see if we are in the correct directory
ls
OUTPUT> yolo_training.ipynb
  • The next set of instructions could be outdated by now

Yolo5

If you with to use Yolo5, you need to clone it from github up to your notebook ( this could become outdated soon) but anyways here are the steps