simet.schemas.transform

simet.schemas.transform

TransformSchema dataclass

TransformSchema(type)

Config schema for selecting a data transform (extensible).

Keeps type as a free string so users can plug in new transforms via a factory/registry without changing the schema. The transform name is later resolved to a concrete implementation that returns a torchvision.transforms.Compose.

Attributes:

Name	Type	Description
type	str	Identifier of the transform to use (e.g., "inception", "my_custom_transform"). The accepted values are defined by your transform registry at runtime.

Example

TransformSchema(type="inception") TransformSchema(type="my_custom_transform")

Notes

• Ensure the chosen transform matches the expectations of your feature extractor/model (e.g., input size, normalization) to avoid feature drift.
• If you later need per-transform options (e.g., image size, mean/std, augmentation flags), extend this schema with optional fields.