From CNNs to YOLO: Understanding Object Detection

Fundamentals Before Training a Steel Defect Detection Model

This article is a learning reflection. I share how I came to understand the workflow of CNNs for image analysis, supported by visualization tools and a few sketches I drew during the learning process.

With this foundation, I then look at how YOLO builds on CNNs to reframe object detection.

Object Detection and Image Analysis

When I first started learning about neural networks, machine learning, PyTorch, etc., my understanding of image analysis was rather vague. I knew that images could be “analyzed” by models, but I did not clearly distinguish between different types of tasks.

Over time, I realized that image analysis involves various tasks, while object detection is one of them, different from classification, image segmentation, etc. Instead of labelling the whole image with a single category, such as “shirt”, “dress”, or “coat”, in image classification, object detection refers to analysing:

Please read the full article in my Medium stories. Here is the link: 👉 From CNNs to YOLO: Understanding Object Detection