Data-Centric Visual AI

2h 4mIntermediate2024-12-09

Authors

Daniel Gural

Course details

This course provides a concise yet comprehensive hands-on experience in implementing data curation methodologies, focusing on creating and refining data feedback loops. By focusing on the iterative data-centric feedback loop, this course helps equip you with the practical skills and mindset needed to build robust and reliable AI systems. Learn how to train computer vision models, critically evaluate their performance, identify weaknesses, and continuously improve your datasets for better results.

Learning objectives
Understand what data-centric visual AI is and why the field is moving away from model-centric.
Learn what makes a good visual AI dataset and where to find it.
Know how to curate a dataset to find mistakes and improve the overall quality.
Develop testing and QA to remove biases and evaluate models.

Skills covered

Neural Networks and Deep LearningArtificial Intelligence FoundationsArtificial Intelligence (AI)One-Off

Concepts

Introduction

Elevate your projects with data-centric visual AI
What is visual AI
What you should know

Introduction to Data Curation for Computer Vision

Data-centric AI paradigm
The feedback loop

Dataset Collection and Visualization

Data sources
GenAI risks to data and an introduction to FiftyOne
Understanding annotations

Data Curation and Improvement

FiftyOne for data analysis
Find image quality issues
Outlier detection and data distribution
Data augmentation

Model Evaluation

Bias detection and mitigation
Model evaluation feedback loop
Advanced model evaluation

Iterative Dataset Improvement and Model Refinement

Iterative dataset improvement
Developing competing models

Conclusion

Connect and continue - Your journey in visual AI