Augmentations are transformations applied to your training images that create new, altered variations of your existing data. By adjusting properties like color, orientation, focus, and framing, each augmentation teaches your model to handle the kind of variation it will encounter in the real world.
Because you can apply several augmentations at once, a single source image can produce many distinct training examples. This multiplies the variety of your dataset without collecting new data, which helps your model generalize better and become more robust to changes in lighting, camera position, and image quality.
The table below lists each available augmentation and links to its dedicated page.
| Augmentation | Description |
|---|---|
| Blur | Add random Gaussian blur to build resilience to camera focus changes. |
| Brightness | Randomly brighten or darken images to handle lighting variation. |
| Camera Gain | Simulate camera sensor gain (ISO) noise. |
| Crop | Randomly crop a subset of each image to vary position and scale. |
| Cutout | Mask portions of an image with black boxes to handle occlusion. |
| Exposure | Adjust gamma exposure to brighten or darken images. |
| Flip | Randomly flip images horizontally or vertically. |
| Grayscale | Convert a percentage of images to grayscale to reduce color reliance. |
| Hue | Shift image hues around the color wheel. |
| Mosaic | Combine multiple images into one tile to improve small-object detection. |
| Motion Blur | Add directional blur that mimics camera or subject motion. |
| Noise | Inject salt-and-pepper noise to handle camera artifacts. |
| 90º Rotate | Rotate images 90 or 180 degrees for orientation invariance. |
| Rotation | Randomly rotate images up to a chosen degree to handle camera roll. |
| Saturation | Adjust color vibrancy to handle saturation differences. |
| Shear | Shear images along their horizontal or vertical axis to vary perspective. |
| </content> |