😶‍🌫️Defects Detection

📝Introduction

This project is a defect detection web UI based on the EBM-YOLOv8 model.

EBM-YOLOv8 = YOLOv8(Backbone) + EMA(Neck) + BiFPN(Neck) + Alpha-MPDIou(Bbox. Loss) + Mish(Activation Function for Convolution)

😎Author

Yinghao He, a graduated student from the North University of China.

🎯How to start

📦Clone the repository:

git clone https://github.com/CherrySama/DefectsDetection.git

🚀Install the required packages:

Pip install the ultralytics package including all requirements in a Python>=3.8 environment with PyTorch>=1.8.

pip install ultralytics
pip install streamlit

🤯Usage:

Dataset:

this project do not have its own dataset, you can write your own dataset yaml cfg and put your dateset in the 'datasets' floder. And if you want to use train.py to train your model, you should revise line12 in train.py to make sure that you could train your model successfully:

    results = model.train(data='VOC_Aluminum.yaml',  # path to your data.yaml
                        epochs=200, 
                        imgsz=640, 
                        workers=8, 
                        batch=16,
                        )

Example usage: yolo train data=VOC_Aluminum.yaml

parent
├── ultralytics
└── datasets

Train the model:

CLI

you can train the YOLOv8 model with a yolo command:

yolo train --data data/defects.yaml --weights yolov8n.pt --epochs 300 --batch-size 16 --device 0 --project runs/train --name defects

yolo can be used for a variety of tasks and modes and accepts additional arguments, i.e. imgsz=640. See the YOLOv8 CLI Docs for examples.

Python

you can also train the YOLOv8 model with a python script:

import torch
from ultralytics import YOLO

model = YOLO("yolov8n.pt")  # load a pretrained model (recommended for training)

model.train(data="data/defects.yaml", epochs=200, batch-size=16, device=0, project="runs/train", name="defects")  # Train the model
metrics = model.val()  # evaluate model performance on the validation set
results = model("https://ultralytics.com/images/bus.jpg")  # predict on an image
path = model.export(format="onnx")  # export the model to ONNX format

or you can simply use the train.py script to train the model:

python train.py 

See YOLOv8 Python Docs for more examples.

🗺️Run the web UI:

streamlit run demo.py

Open the web browser and visit http://localhost:8501 to start the web UI.

😮‍💨Last words

This is my little graduation project, I'll be glad that you could use it as your baseline model or just simply have a look.
If you have any questions or suggestions, please feel free to contact me at the following Email address 925782272@qq.com or create a new issue.