Module 8: Object Detection

1. Learning Objectives

Understand the basics of object detection
Learn the differences between detection and classification
Explore region-based and single-shot detection methods

2. Object Detection vs. Classification

Classification: Predicts a single label for the entire image.

Detection: Identifies and locates multiple objects with bounding boxes and labels.

3. Key Metrics

IoU (Intersection over Union): Measures overlap between predicted and ground truth boxes.
mAP (mean Average Precision): Standard metric for evaluation across classes.

4. Object Detection Models

R-CNN: Region proposals + CNN classifier (slow but accurate)
Fast R-CNN: Faster version using shared features
Faster R-CNN: Uses Region Proposal Network (RPN)
YOLO (You Only Look Once): Real-time detection using a single network
SSD (Single Shot MultiBox Detector): Faster with multi-scale feature maps

5. Anchor Boxes and NMS

Anchor Boxes: Predefined bounding boxes of different aspect ratios
Non-Maximum Suppression (NMS): Removes redundant overlapping boxes

6. Example: YOLOv5 Inference


# Example using YOLOv5
!git clone https://github.com/ultralytics/yolov5
%cd yolov5
!pip install -r requirements.txt

# Run inference
!python detect.py --source ../your_image.jpg --weights yolov5s.pt --conf 0.5

7. Assignment

Objective: Understand and experiment with object detection.

1. Run YOLOv5 on your own image and analyze the results
2. Compare YOLO with a Faster R-CNN model (optional)
3. Report bounding box results and detection confidence
4. Write a 1–2 page report about performance, speed, and accuracy

Due: End of Week 8