Visualize Augmentation¶

Before training, it's helpful to preview the effects of your augmentation pipeline. This visualize-augmentation CLI script lets you quickly visualize how your images will look after transformation, giving you an intuitive sense of whether the augmentations are too strong, too weak, or just right.

Image augmentations affect model generalization, so it's important to visually inspect them to ensure they're appropriate and realistic.

Example: Side-by-Side Comparison¶

fast-plate-ocr visualize-augmentation \
  --img-dir benchmark/imgs \
  --columns 2 \
  --rows 4 \
  --show-original \
  --augmentation-path transforms.yaml \
  --plate-config-file config/latin_plates.yaml

This shows 8 images (2x4 grid), each combining the original (left) and the augmented (right) version.

Augmented Images

Default Augmentation¶

If --augmentation-path is not provided, the tool uses the default training pipeline from fast_plate_ocr.train.data.augmentation.

RGB/Grayscale default augmentation

When using rgb in the plate config, the default augmentation applied is slightly different than the one when choosing grayscale (it basically has more color/noise transforms).