
Welcome to the course! In this introductory lecture, I'll give you an overview of what we'll build and learn throughout the course.
You'll discover how we'll combine YOLO26, Python, OpenCV, Gemini, Generative AI, and AI Vision Agents to build practical Computer Vision applications.
We'll cover the major topics and hands-on projects you'll work on, including:
YOLO26 for object detection, segmentation, pose estimation, classification, and OBB
Custom dataset creation and YOLO26 model training
Real-time Computer Vision applications
Vehicle intensity heatmaps and Bird's Eye View (BEV) systems
AI Vision Agents using LLMs, Gemini, and YOLO
OCR, Document AI, and invoice processing
Food and Nutrition Vision applications
By the end of the course, you'll have a clear understanding of the learning path and the practical projects we'll build together.
Let's get started!
In this lecture, you will learn how to install and set up uv, an ultra-fast Python package manager written in Rust. We will cover step-by-step installation instructions for Windows, macOS, and Linux, and demonstrate how to verify the setup. Finally, we will run a quick hands-on test to initialize a project, add dependencies, and run Python code safely within a managed environment.
Learn how to quickly generate your own Google Gemini API Key through Google AI Studio, so you can start building AI-powered applications right away.
In this video tutorial, we will explore how to perform Object Detection, Instance Segmentation, Pose Estimation, Image Classification, and Oriented Bounding Box (OBB) Object Detection using YOLOv8, YOLO11 & YOLO26 on Images and Videos. We will also explore YOLOE-26 for Open Vocabulary Instance Segmentation in detail.
In this lecture, you will learn how to annotate and label a custom dataset using Roboflow for object detection tasks. We will cover image upload, creating classes, drawing bounding boxes, organizing data, and exporting the dataset in YOLO format. By the end of this lecture, you will have a fully prepared dataset ready for training your YOLO26 object detection model.
In this lecture, you will learn how to train a YOLO26 object detection model on a custom pothole dataset. We will cover dataset configuration, training setup, model parameters, and performance monitoring. You will also learn how to evaluate the trained model and run inference to detect potholes in new images and videos.
Learn to create real-time vehicle intensity heatmaps from YOLO26 detections. Filter vehicles, map traffic intensity, apply blur and color maps, and overlay on video for traffic visualization.
In this lecture, you will learn how to transform standard camera-view detections into a real-time Bird’s Eye View (BEV) system using YOLO26 and OpenCV. We will start by selecting accurate road perspective points from a video frame.
You will then project detected vehicle positions into a top-view plane, allowing you to visualize traffic from an overhead perspective. By the end of this lecture, you will have built a complete BEV mapping system that converts object detection outputs into a structured and analytical top-view representation—ready for applications like traffic monitoring, counting, and heatmap generation.
In this tutorial, we will build a Vision Agent — a minimal agent application using Streamlit. Unlike a full autonomous agent, this is a minimal agent: the LLM sits at a decision point, choosing between tools and adapting based on what it observes, with only a single retry loop (not an open-ended plan-act-observe cycle) and no memory across requests.
We will build a Streamlit application where users can upload an image and ask a question, such as "count the cars in the image" or "count the people in the image." The agent is equipped with a single computer vision tool — a YOLO Object Detection tool — which it uses to answer counting-based questions.
If the YOLO tool is unable to answer a given question, the agent falls back to Gemini 3.1 Flash-Lite (a Vision-Language Model) to generate a response instead.
By the end of this lecture, you'll understand how to connect an LLM to a CV tool for simple visual reasoning tasks, and how to add a fallback mechanism using a VLM when the primary tool falls short. Note: this pipeline currently handles single-object questions only (e.g., it cannot yet handle combined queries like "count the cars and people" in one request).
In this lecture, we build a Vision Agent — a fully autonomous computer vision application powered by Gemini and YOLO. Rather than hard-coding a fixed pipeline, we design an AI agent that decides for itself, in a multi-step loop, whether to call the YOLO object detection tool.
You'll also learn about Gemini's native function calling, which lets an LLM control a multi-step reasoning loop with zero manual JSON parsing. The application can handle multi-object questions (e.g., "count the cars and persons in the image") and is built to support multiple tools — though in this tutorial we give the agent a single tool, YOLO object detection. In the next tutorial, we'll expand the agent to use multiple tools.
Throughout, the agent decides for itself, in a multi-step loop, when it has gathered enough information to give a final answer.
In this tutorial, we give the agent multiple tools, and it learns to choose between them based on the question: the YOLO object detection tool for counting, the YOLO segmentation tool for area and shape, and an OCR tool for reading text.
For OCR, we use Gemini 3.1 Flash Lite — a multimodal vision-language model that accepts both text and images as input and can read and extract text.
For segmentation, we use YOLO26 segmentation model weights to perform pixel-level segmentation — drawing a mask around each detected object to get both an exact count and the percentage of the image area it covers.
In this lecture, we build a Document Agent — an autonomous AI application that processes receipts and invoices from an image. The agent decides for itself which tools to use based on the question: it can read and transcribe the raw text (OCR), extract structured details like vendor name, line items, subtotal, tax, discount, and total, and verify whether those numbers actually add up correctly.
You'll see how a single agent can combine multiple tools — including a plain Python calculation tool, not just another AI call — to check whether an image document's numbers genuinely add up, correctly accounting for tax and discounts along the way. This lecture brings together everything from the earlier Vision Agent lectures into one practical, real-world application you can extend and add to your portfolio.
In this lecture, we build a Food & Nutrition Agent — an autonomous AI application that identifies food in a photo and looks up real nutrition information about it. The agent decides for itself which tools to use: it detects common food items with YOLO, reasons visually about food YOLO can't classify, and looks up real calorie, protein, fat, and carb data from the USDA FoodData Central database — a live external API, not the LLM's own memory.
This lecture introduces a new kind of tool: instead of another vision or AI model, the agent calls a real external API to fetch verified, factual data. You'll see why this matters — grounding answers in a real data source instead of letting the LLM guess helps avoid inaccurate nutrition claims, and this same pattern (agent + external API) can be reused for countless other real-world applications.
Learn how to build practical AI-powered Computer Vision applications using YOLO26, Gemini, and AI Vision Agents.
In this course, you’ll learn YOLO26 for object detection, segmentation, pose estimation, classification, and OBB, build AI Vision Agents with Gemini and YOLO, and create real-world applications for OCR, Document AI, invoice processing, and food & nutrition analysis.
The course focuses on practical projects that combine Computer Vision, Generative AI, and Vision Agents.
What You Will Learn
YOLO26 & Computer Vision
YOLOv8, YOLO11 & YOLO26
Object Detection, Segmentation, Pose Estimation, Classification & OBB
YOLO26 Custom Object Detection: Dataset Creation & Custom Model Training
How to Annotate / Label a Custom Dataset Using Roboflow
Train YOLO26 on a Custom Pothole Dataset for Pothole Detection
Advanced Real-World Projects
Building a Vehicle Intensity Heatmap from YOLO26 Detections
Real-Time Bird’s Eye View (BEV) System using YOLO26 and OpenCV
AI Vision Agents for Computer Vision
Build a Minimal Vision Agent using LLM + YOLO
Object Counting with YOLO and Streamlit
Autonomous Vision Agents with Gemini & YOLO
Build a Fully Autonomous AI Vision Agent with Gemini & YOLO
Add Segmentation and OCR tools
Document & Invoice Processing
Build a Document & Invoice Processing Agent
Extract and analyze information from documents
Food & Nutrition Vision Agent
Build a Food & Nutrition Agent
Combine Computer Vision with nutrition lookup using real USDA data