Welcome to the first course in the Edge AI and Robotics basket. In this course, you will learn how to think like an AI engineer by building real-world computer vision applications rather than simply studying the underlying theory.
Imagine you are asked to develop a live face attendance system for your classroom. Where would you begin? How would you decide which AI techniques to use? What challenges would you face, and how would you improve the system? By the end of this course, you will be able to answer these questions with confidence.
You will begin by learning the three fundamental building blocks of almost every computer vision application:
Image Classification – teaching a computer to recognize what is present in an image.
Object Detection – identifying and locating multiple objects within an image.
Image Segmentation – understanding objects at the pixel level for detailed scene analysis.
Using these concepts, you will build complete AI applications and understand how they work as a whole. Initially, you will treat the system as a black box, focusing on assembling and deploying working solutions. Once you are comfortable building complete applications, you will gradually open that black box and explore the techniques that make these systems accurate and efficient. You will study concepts such as optimization, regularization, model improvement, and other practical methods used in modern AI development.
The course is designed around interactive Streamlit-based learning modules along with separate lab components for each module, allowing you to experiment with models, visualize their behavior, and gain hands-on experience. Alongside these activities, detailed documentation and reading material will help strengthen your understanding.
Unlike traditional theory-heavy courses, this course emphasizes real-world deployment. Our philosophy is simple:
Learn how to build it.
Learn how to make it better.
By the end of the course, you will not only understand the core ideas behind computer vision but also develop the confidence to design, build, and improve practical AI systems for real-world applications.
Mid Semester Test (MST): 25 Marks [Syllabus: Module 1-5, Refer to some class notes added in the course git, study guide and practice questions on LMS)
End Semester Test (EST): 45 Marks
Sessional: 30 Marks
Group Project 1: 10 Marks
Group Project 2: 10 Marks
Laboratory Performance: 10 Marks
During your weekly practical sessions, you will complete lab assignments. Each lab carries 4 marks, and your overall laboratory performance will be scaled to 10 marks as part of the sessional evaluation.
We are excited to have you on this journey. Let's start building intelligent computer vision systems!
Task Identification
Specifics Identification
Learning Outcome: Learn how to analyze a real-world problem and identify the appropriate computer vision solution.
Build a Model: Classification
Build a Model: Transfer Learning
Learning Outcome: Build complete image classification systems using both custom and pretrained models.
Dive into Model Design
Building Blocks
Parameters
Tensor Shapes
Advanced Architectures
Learning Outcome: Understand how modern vision networks are designed and why different architectures perform differently.
Dive into Data Augmentation
Learning Outcome: Learn how data quality and augmentation techniques improve model performance and generalization.
Dive into Optimization
Dive into Loss Functions & Metrics
Activation Functions
Learning Outcome: Understand how neural networks learn, how they are optimized, and how their performance is measured.
Image Segmentation
Object Detection
Learning Outcome: Extend your knowledge from image classification to localization and pixel-level understanding.
Pruning
Quantization
Memory & Compute Optimization
Learning Outcome: Learn techniques to compress and optimize deep learning models for deployment on edge devices.
Regularization
Image Normalization
Feature Normalization
Weight Initialization
Learning Outcome: Explore techniques that improve model stability, convergence, and generalization.
Generative Adversarial Networks (GANs)
Diffusion Models
Learning Outcome: Learn the foundations of modern image generation techniques and understand how generative models create realistic visual content.