Deep Learning Explained: Computer Vision

This repository contains small tutorials and reference projects around the basic concepts of modern Computer Vision using Deep Learning. A high-level overview of the topics covered and the respective ideal learning outcomes in this course is provided below.

Course Content

Introduction to Computer Vision and Deep Learning Basics

• Gain a fundamental understanding of computer vision and its intersection with deep learning.
• Learn how to create, annotate, and manipulate image datasets.
• Explore the basics of image classification using a simple linear layer.
• Observe how Convolutional Neural Networks (CNNs) interpret image data.
• Get introduced to the concept of data augmentation for enhancing performance of computer vision models.

Jupyter Notebook: basics/introduction.ipynb

Getting Started with Image Classification

1. Image Preprocessing: Explore techniques to prepare and preprocess image data for model training.
2. Building CNN Models: Learn how to build your first Deep Learning model for image classification with Tensorflow and Keras.
3. Train your CNN model: Train your first CNN model, explore different training callbacks, and understand the role of the loss function.
4. Evaluating Model Performance: Learn how to evaluate the performance of image classification models using metrics like accuracy, precision, and recall and visualize the confusion matrix.
5. Test performance on unseen data: Test the performance of your model on data unkown to the model.

Jupyter Notebook: image_classifier/classification.ipynb

Object Detection with YoloV8

1. Introduction to Object Detection: Dive into the advanced Computer Vision task of object detection by training your first state-of-the-art model.
2. Introduction to YoloV8: Familiarize yourself with the YoloV8 architecture and it's implementation in Keras using the Keras-CV API.
3. Object Detection data pre-processing: Get hands-on experience on how to prepare a dataset to train an object detection Deep Learning model.
4. Training YoloV8: Experiment with different training configurations and strategies to optimize the YoloV8 model's performance.
5. Evaluating Object Detection Performance: Understand evaluation metrics for object detection, such as mean Average Precision (mAP).
6. Test performance on unseen data: Test the performance of your model on data unkown to the model.

Jupyter Notebook: object_detector/object_detection.ipynb

Semantic Segmentation with UNet

1. Introduction to Semantic Segmentation: Explore the advanced Computer Vision task of semantic segmentation.
2. UNet Architecture: Understand the U-Net architecture and it's implementation with the Keras Functional API.
3. Data Preparation: Learn how to prepare a dataset of images and segmentation mask annotations for a semantic segmentation model.
4. Transfer Learning: Understand the concept of Transfer Learning by leveraging a pre-trained feature extraction model as basis to train your model.
5. Training and Evaluation: Train the UNet model, explore more training callback functions and evaluate its performance using metrics like Intersection over Union (IoU).

Jupyter Notebook: semantic_segmentation/semantic_segmentation.ipynb

Use AWS Sagemaker to train your model in the cloud

Familiarize yourself with the AWS environment: Familiarize yourself with tasks like data storing on S3 buckets, the concepts of users,roles and policies in AWS and learn how to set up and run a model training job using AWS Sagemaker.

NOTE: Before running any of the following scripts make sure to add your authentication credentials and following the instructions provided in each script.

NOTE: Remember to delete resources and keep track of how much you are charged for anything related to AWS Services, as extensively using cloud resources can be quite costly!

Upload dataset to an S3 Bucket script: sagemaker/upload_dataset.py

Training job entrypoint script: sagemaker/train.py

Train a deep learning model with Sagemaker script: sagemaker/run_sagemaker_job.py

Extras

Utility functions for all modules: helper.py
Google Collaboratory supported notebooks: google_collab/