← Computer Vision learning pathComputer Vision · Module 9 of 10
Data, labelling and augmentation
Prepare high-quality image datasets.
📖 Key ideas
- Good data matters more than fancy models.
- Label tools: CVAT, Label Studio, Roboflow.
- Augmentation: flips, crops, colour jitter, mosaic (Albumentations).
- Split train/validation/test without leakage (no same scene in both).
- Check label quality and class balance.
💡 Real-life example: Labelling data is teaching a child with clear picture cards; wrong cards teach wrong lessons.
AnnotationAdding labels to images
AlbumentationsImage augmentation library
LeakageTest data too similar to training
Class balanceSimilar examples per class