Train-time image augmentation creates new training images by randomly transforming existing images, thereby increasing the size of (i.e., “augmenting”) the training data. This allows projects to be built with datasets that might otherwise be too small. In addition, all image projects that use augmentation have the potential for smaller overall loss by improving the generalization of models on unseen data.
If you add a secondary dataset with images to a primary tabular dataset, the augmentation options described above are not available. Instead, if you have access to Composable ML, you can modify each needed blueprint by adding an image augmentation vertex directly after the raw image input (as the first vertex in the image branch) and configure augmentation from there.
Set transformations prior to modeling¶
After selecting your target, toggle on the Image Augmentation tab in Advanced options.
From there, begin selecting transformation settings, described here. These settings will be applied to all models when running Autopilot or using the Repository.
You can continue to modify settings, clicking Preview augmentation to view a sample of results:
The settings you choose are automatically saved as a list named Initial Augmentation List. If you do not set transformations through Advanced options, you can later create augmentation lists using the Advanced Tuning tab.