Computer Vision Section 066
Vision Datasets and Annotation
The unglamorous work that decides your accuracy: labelling images well, and finding the errors already in your data.
11 of 11 lessons published Three reading levels on every lesson
Lessons in order
Work top to bottom. Each lesson assumes the one above it.
- Choosing an annotation tool
- Writing annotation guidelines
- The COCO dataset format
- Converting between dataset formats
- Building an image dataset from the web
- Finding label errors in an image dataset
- Active learning for labelling
- Model-assisted labelling
- Deduplicating an image dataset
- Data leakage in image splits
- Dataset bias and shortcut learning