iss-egypt-academic/YOLO_Workshop

YOLO workshop that was operated by ISS Egypt @ Universiti Teknologi Malaysia, Johor Bahru # YOLO_Workshop YOLO workshop that was operated by ISS Egypt @ Universiti Teknologi Malaysia, Johor Bahru # INTRODUCTION You Only Look Once is an effiecient deep learning based object detection model. It predicts bounding boxes and and class propabilities in a single forward pass of the network This makes it incredibly fast and suitable for real-time applications. In this workshop, you will: Collect and preprocess data Train a YOLO model Evaluate and visualize predictions Run YOLO for real-time object detection # Requirements To follow along with this workshop, you need: Python 3.7+ Jupyter Notebook or Google Colab OpenCV TensorFlow / PyTorch YOLOv5 or YOLOv8 framework You can intall dependencies using: ``` pip install -r requirements.txt ``` # Workshop Steps ## Step 1: Gathering Data Capture images using a webcam or use an existing dataset. Save images into a folder for annotation. Example function to capture images in Google Colab: ``` from IPython.display import display, Javascript from google.colab.output import eval_js from base64 import b64decode import os // Function to access webcam and capture images def take_photos(num_photos=3, folder_name='captured_photos'): # Implementation... ``` ## Step 2: Preparing the Dataset Annotate images using LabelImg or Roboflow. Convert annotations to YOLO format. Split the dataset into training and validation sets. ## Step 3: Training the YOLO Model Download and configure YOLOv5 or YOLOv8. Train the model on annotated data. Example for YOLO5 training ``` python train.py --img 640 --batch 16 --epochs 50 --data dataset.yaml --weights yolov5s.pt ``` ## Step 4: Evaluating the Model Analyze training metrics (mAP, loss, precision, recall). Visualize predictions on test images. ## Step 5: Running Object detection Perform inference using the trained YOLO model. ``` python detect.py --weights runs/train/exp/weights/best.pt --source test_images/ ```