← AI Terminology
Object Detection
Object detection is a computer vision task that simultaneously identifies what objects are present in an image and where they are — producing bounding boxes with class labels and confidence scores for each detected object instance.
It combines image classification (what) with localisation (where).
It combines image classification (what) with localisation (where).
Why It Matters in AI
Object detection is the core perception task for autonomous vehicles (detect cars, pedestrians, traffic signs), surveillance (detect persons, weapons), medical imaging (detect tumours, lesions), and industrial quality control (detect defects). It evolved from slow R-CNN pipelines (2014) to real-time single-shot models (YOLO) that can detect 80+ object classes at 60 FPS on a GPU — enabling applications previously impossible in real time.
Key Points
| Aspect | Description |
|---|---|
| YOLO | You Only Look Once — fastest practical detection family; YOLOv8/v11 are current state-of-art |
| Output | Set of (class, confidence, x1, y1, x2, y2) bounding boxes per image |
| One-stage | YOLO, SSD, RetinaNet: predict all boxes in a single forward pass — real-time capable |
| Two-stage | Faster R-CNN: region proposal → classification — high accuracy, slower |
| Evaluation | mAP (mean Average Precision) at IoU thresholds — COCO mAP@0.5:0.95 is the standard |
| Anchor-free | FCOS, CenterNet, DETR — no pre-defined anchor boxes; predict object centres or use attention |
Simple Analogy
Looking at a crowded photo and drawing boxes around every face with a name tag attached: that is object detection. Finding the face is localisation; naming it is classification. The challenge is doing both simultaneously, for dozens of objects, in milliseconds.
Common Usage Examples
from ultralytics import YOLO; model = YOLO("yolov8n.pt"); results = model("image.jpg")results[0].boxes.xyxy— bounding box coordinates;.cls— class IDs;.conf— confidence scores- Faster R-CNN:
from torchvision.models.detection import fasterrcnn_resnet50_fpn - COCO dataset: 80 categories, 330K images — standard detection training and evaluation benchmark
- Real-time: YOLOv8n at 640px: ~160 FPS on RTX 3080 — autonomous vehicle perception stack
Summary
In short: Object detection finds and locates every object in an image simultaneously — the foundational computer vision task powering autonomous vehicles, surveillance, medical imaging, and any application requiring real-time scene understanding.