1)Create a Clove of This Repository
2)Create a Dataset Folder of images(png/jpg) and labels(jpg) to it in YOLOv5 Format
YOLOv5 Format is shown Below
The Folder Structure of Dataser Folder Should be like
3)After Placing the Images and Labels in the Dataset Folder
Edit the Config.Yaml file in our Repository
As in above image we can see the File contains 4 parameters to be changed 1)Training images Path:This Should be something like "Project/Dataset/images/train" 1)Validation images Path:This Should be something like "Project/Dataset/images/validation" 3)Number of Classes:This Determines how many types of classes or objects we want to identify using YOLOv5 4)Names of Classes:This Contains names of all classes
4)After Doing this We need to install all the Required Modules used by YOLOv5 for its proper working using pip The Command to be used in CLI to install all Modelues while being in our Project Directory will be: pip install -r requirements.txt
5)Now For Training The Model with your dataset just use this command: python train.py --img 416 --batch 8 --epochs 100 --data config.yaml --cfg models/yolov5s.yaml --name run_name
In the above image the img value 416 Depends on the dimensions of the images you are using to train and validate and name can be anything you like and all other parameters can also be changed.
6)After having a successful run of the above command we will have a new Folder named runs in which there will be a folder named same as the run_name you gave above in the command Go in that folder and then go in Weights folder and copy the best.pt file in our main Project Directory
This File will be used to Detect the Hats(Objects)
7)Now using that Weights File we will run this Command python detect.py --source "Path of the images in which you want to detect the Objects" --weights best.pt
8)After Running this Command The Output images will be again in the runs Folder
The Diference Between Before Detection and After Can be seen below:
This Project is based on YOLOv5 by https://github.com/ultralytics/yolov5 Huge Thanks to Them For making it so easy...