Build Your Own AI Waste Sorter: A YOLOv10 Course

Every year, the world generates over 2 billion tonnes of waste. Much of it ends up polluting land, oceans, and communities. But what if technology could help us sort, understand, and reduce that waste automatically?

Thanks to computer vision and AI, we're making real progress. Apps like DWaste already help users identify waste types like plastic and paper, building smarter recycling habits. In this course, you'll learn to build your own waste detection system from scratch using YOLOv10.

Teaching machines to recognize waste

What You'll Build

A complete object detection system that can identify seven types of waste: biological, cardboard, glass, metal, paper, plastic, and trash. You'll go from raw images to a working model you can deploy on the web.

Course Breakdown

1. Dataset Preparation

We'll use the Garbage Classification v2 dataset from Kaggle, which contains over 19,000 images across ten classes. For this project, we focus on seven categories most relevant to waste sorting. You'll learn to extract, organize, and rename images for annotation.

2. Annotation with Annotate-Lab

Before any model can detect waste, it needs to learn what waste looks like. You'll set up Annotate-Lab, an open-source annotation tool, and label images with bounding boxes. Each annotation includes class IDs from 0 to 6, and you'll export them in YOLO format—one text file per image with normalized coordinates.

3. Exploratory Data Analysis

Understanding your data is essential. You'll analyze your annotated dataset to check class distribution, visualize bounding boxes, and verify that every image has a matching label file. This step catches errors before training begins.

4. Data Augmentation

Real-world conditions vary. You'll apply transformations like rotation, flipping, brightness adjustments, and color tweaks to create five augmented versions of each image. Your script will automatically update the bounding boxes to match, boosting model generalization.

5. Image Classification vs. Object Detection

Classification answers "what is it?" with a single label. Detection answers "what is it and where is it?"—essential for waste images that often contain multiple items. That's why we use YOLO.

6. Training YOLOv10

Using the Ultralytics YOLO framework, you'll train a model on your annotated dataset. With a simple command, YOLOv10 learns to detect and classify waste. You'll monitor precision, recall, and mAP metrics as training progresses.

7. Testing and Export

Once training is complete, you'll test your model on new images and see bounding boxes drawn around detected waste. You'll also learn to export your model to formats like ONNX or TensorRT for deployment on mobile or edge devices.

Ready to Build?

This course gives you everything you need to create an AI waste sorter that actually works. No prior experience required—just curiosity and a willingness to learn.

📺 Watch the full course here

Cheers, and happy building! 😊

Sources

• Garbage Classification v2 Dataset: Kaggle
• Ultralytics YOLOv10 Documentation: docs.ultralytics.com
• World Bank. (2018). What a Waste 2.0: openknowledge.worldbank.org