Deep-Learning-Projects-Computer-Vision

It is a repository, in which I upload Projects about Computer Vision

---

Computer Vision for Deep Learning - Neuralearn Course

This repository contains projects and notes from my "Computer Vision for Deep Learning" course, taken from Neuralearn. The course provided a solid foundation in deep learning concepts and practical experience with modern computer vision techniques. Below is an overview of the topics covered and projects I worked on.

Course Overview

TensorFlow and Tensors:

• Introduction to TensorFlow and its core concepts.
• Working with tensors, ragged tensors, and performing tensor operations.

Convolutional Neural Networks (CNNs):

• Deep dive into CNN architecture, covering layers like convolution, pooling, and activation.
• Implemented CNNs for various projects, including image classification and detection tasks.

Deep Learning Architectures:

• Explored popular deep learning models, including:
  • LeNet
  • MobileNet
  • ResNet
  • EfficientNet
  • VGG
  • AlexNet
• Compared performance and use cases for different architectures.

Projects:

1. Car Price Prediction:

   • Built and trained a model to predict car prices using regression techniques.

2. Malaria Disease Classification:

   • Data collection, preprocessing, and visualization.
   • Created a deep learning model to classify malaria-infected and healthy cells.
   • Applied techniques to handle overfitting, data augmentation, and used custom layers/models (Keras API, functional, and subclassing).

3. Human Emotion Detection:

   • Developed a deep learning model to detect human emotions from images.
   • Techniques included transfer learning, fine-tuning, and Hugging Face transformer tuning.
   • Performed black-box testing (visualizing convolution layers) and white-box testing (comparing predictions with original samples).

Transformers:

• Explored transformer models for tasks like image classification and generation.
• Implemented fine-tuning for specific applications using pre-trained transformer models.

MLOps:

• Focused on managing machine learning workflows with:
  • Dataset and model versioning.
  • Experiment tracking using Wandb.
  • Model testing and deployment on the cloud.

Quantization:

• Explored post-training quantization and Quantization-Aware Training (QAT) for optimizing model performance on edge devices.

YOLO v2 (You Only Look Once) Architecture:

• Theoretical and practical exploration of the YOLO v2 object detection architecture.
• Implemented custom loss functions, Non-Max Suppression, and Intersection over Union (IoU).

Image Generation and Generative Models:

• Studied image generation techniques using:
  • Variational Autoencoders (VAE)
  • Generative Adversarial Networks (GANs)
• Projects included generating handwritten digits (MNIST) with VAE and generating realistic celebrity faces using DCGAN and the CelebA dataset.

Key Learnings

This course provided in-depth knowledge in both foundational and advanced computer vision topics, allowing me to apply theoretical concepts through hands-on projects. I gained experience with real-world workflows in machine learning, from data collection and model building to deployment and MLOps.

Feel free to explore the projects and code in this repository for more insights!

---