Explore a powerful visualization technique for vision classification in this 50-minute tutorial. Learn how to transform images into embeddings, project them using UMAP, and create dynamic visualizations. Discover the process of using Img2Vec to obtain embeddings, reducing dimensionality with UMAP, and creating interactive visualizations with Bokeh. Follow along as the instructor demonstrates saving PNG thumbnails in DataFrames, integrating HTML into notebooks, and analyzing the resulting visualizations. Gain insights into clustering high-dimensional image data and understanding relationships between images through this hands-on project focused on aerospace engine defect detection.
Overview
Syllabus
- Introduction:
- Project Overview:
- Data Overview:
- Getting Embeddings with Img2Vec:
- Reducing Dimensionality with UMAP:
- Interactive Visualization with Bokeh:
- Saving PNG Thumbnail in DataFrame:
- Adding Ugly HTML into the Notebook:
- Analyzing the Visualization:
- Conclusion:
Taught by
Yacine Mahdid
