Lab notebook · 09 / 27Computer vision

Object detection with YOLOv8

What does a pretrained detector find in one image, and what remains after a confidence filter?

Self-directed study

Scope of this result

A one-image inference demonstration. Detection counts describe this image and threshold, not detector accuracy.

The evidence

Fig. 01

Street scene beside a bus with five YOLOv8n detection boxes: one bus and four people, each labelled with a confidence score. — YOLOv8n predictions at the declared 0.25 inference threshold: one bus and four people.

01 · Method and interpretation

A reproducible COCO-pretrained YOLOv8n inference demonstration on an official Ultralytics image pinned by Git commit and SHA-256. One declared pass finds five boxes—one bus and four people—then a transparent filter keeps three people at confidence 0.50 or higher. These counts describe one image and threshold, not detector accuracy; a real evaluation needs representative labelled images and precision, recall, or mean average precision.

02 · Supporting figures

Selected outputs from the complete notebook.

Fig. 02

The same street scene with green boxes around three people whose YOLOv8n confidence scores are at least 0.50. — A class-and-confidence filter retains the three people scored at least 0.50.