Undergraduate Project - Computer Vision

Overview

This undergraduate project focuses on exploring and developing advanced techniques in computer vision, structured into two main sections: Classical Computer Vision and Modern Deep Learning-based Computer Vision. The project begins by covering the fundamentals of image processing, video analysis, and classical object detection algorithms. In the second part, using cutting-edge tools like PyTorch, convolutional neural networks (CNN) will be implemented. The project also delves into modern techniques such as transfer learning, pre-trained models, and the emerging Vision Transformers. Furthermore, efficient object detection with YOLO and facial recognition using deep learning will be thoroughly explored.

Table of Contents

1. Project Structure
2. Modern CV vs Classical CV
3. Roadmap
4. Requirements
5. References

Project Structure

The project will be organized into the following sections:

1. Classical Computer Vision

• Image Processing

  • Image processing

• Video Processing

  • Working with Video

• Object Detection

  • Object Detection

• Face Detection

  • Face Detection

• OCR

  • Text Detection

2. Modern Computer Vision - Deep Learning

• Image Classification

  • Cnn Model theory

  • Cnn With Keras (Normal, Pneumonia and Tuberculosis Lungs)

  • CNN With Pytorch Lightning (Fundus Glaucoma Detection)

  • Residual Network With Pytorch Lightning (Normal, Pneumonia and Tuberculosis Lungs)

• Image Manipulation

  • AutoEncoder With Keras (The main objective of creating this dataset is to create autoencoder network that can colorized grayscale landscape images)

• Image Generation

  • DCGAN With Keras (GAN using the MNIST dataset)

• Image Segmentation

  • Lungs Segmentation Using U-Net architecture

  • Tumor Segmentation Using Yolo v11

• Object Detection

  • Object Detection Using YOLO v5 (road-detection)

  • Object Detection Using YOLO v8 (Bone Fracture Detection)

  • Object Detection Using YOLO v11 (Tumor Detection)

• Object Tracking

  • Object Tracking With YOLO v11 - Download Result Video

  • Tracking and Counting With YOLO v11 - Download Result Video

  • Speed Estimation With YOLO v11 - Download Result Video

  • Heat Maps With Yolo v11 - Download Result Video

• Pose Estimation

  • Pose Estimation With Yolo v11 (human pose estimation)

• Face Recognition

  • Facial Recognition With Deep Face

Modern CV vs Classical CV

Topic	Classical Computer Vision	Modern Computer Vision (Using Deep Learning)
Definition and Core Principles	Based on hand-crafted algorithms and engineered features.	Based on deep neural networks and learning from large datasets.
Feature Extraction Method	Features are extracted manually using algorithms like SIFT, SURF, and HOG.	Features are automatically learned by convolutional neural networks (CNNs).
Accuracy	Accuracy is usually limited and optimization for complex problems is harder.	Generally achieves much higher accuracy, especially in complex tasks like image recognition and classification.
Amount of Data Required	Requires less data, but models are usually less optimized.	Requires large amounts of data for effective learning.
Computational Complexity	Generally lighter and simpler in computation.	Heavier in computation and requires powerful hardware (GPUs).

Roadmap

The project will progress through the following phases:

1. Phase 1: Implementing Classical Computer Vision Techniques
2. Phase 2: Building and Training Deep Learning Models with PyTorch and tensorflow, keras

Requirements

To run the project, you'll need:

• Python 3.x
• tools: Jupyter Lab, Colab, lightning studio
• Required libraries listed in requirements.txt.

You can install the dependencies using:

pip install -r requirements.txt

References:

• Ayyadevara, V. K., & Reddy, Y. (2024). Modern Computer Vision with PyTorch - Second Edition: A practical roadmap from deep learning fundamentals to advanced applications and Generative AI (2nd ed.). Packt Publishing.
• Elgendy, M. (2020). Deep Learning for Vision Systems (1st ed.). Manning.
• Ratan, R. D. (2024). Modern Computer Vision GPT, PyTorch, Keras, OpenCV4 in 2024! Next-Gen Computer Vision: YOLOv8, DINO-GPT4V, OpenCV4, Face Recognition, GenerativeAI, Diffusion Models & Transformers [Online course]. Udemy.