Schema Reference¶

Every column type in 3LC is defined by a Schema. The convenience schemas below are the recommended way to declare columns — each one configures the right data type, storage, and Dashboard features automatically. The table lists each convenience schema, the Python type it produces in sample view, how the data is stored, and where to read more.

Schema	Python type	Storage	Details
`ImageSchema()`	`PIL.Image`	file (PNG)	Images
`ImageSchema(sample_type=None)`	`str` (URL)	inline	Images
`CategoricalLabelSchema(classes=)`	`int`	inline	Categorical
`BoundingBoxes2D.schema(classes=)`	`BoundingBoxes2D`	inline	Bounding Boxes
`OrientedBoundingBoxes2D.schema(classes=)`	`OrientedBoundingBoxes2D`	inline	OBBs
`Keypoints2D.schema(classes=, num_keypoints=)`	`Keypoints2D`	inline	Keypoints
`SegmentationPolygons.schema(classes=)`	`SegmentationPolygons`	inline	Instance Seg.
`SegmentationMasks.schema(classes=)`	`SegmentationMasks`	inline	Instance Seg.
`SemanticSegmentationSchema(classes=)`	`PIL.Image`	file (PNG)	Semantic Seg.
`Float32Schema(shape=, sample_type="numpy_array")`	`numpy.ndarray`	inline	Embeddings
`Float32Schema(shape=, sample_type="torch_tensor")`	`torch.Tensor`	inline	Embeddings
`Float32Schema()`	`float`	inline	Tabular
`Float64Schema()`	`float`	inline	Tabular
`Int8Schema()`	`int`	inline	Tabular
`Int16Schema()`	`int`	inline	Tabular
`Int32Schema()`	`int`	inline	Tabular
`Int64Schema()`	`int`	inline	Tabular
`Uint8Schema()`	`int`	inline	Tabular
`Uint16Schema()`	`int`	inline	Tabular
`Uint32Schema()`	`int`	inline	Tabular
`Uint64Schema()`	`int`	inline	Tabular
`BoolSchema()`	`bool`	inline	Tabular
`StringSchema()`	`str`	inline	Tabular
`UrlSchema()`	`str` (URL)	inline	Tabular
`DatetimeStringSchema()`	`str` (datetime)	inline	Tabular

For large arrays that should be stored as files rather than inline, see ExternalNumpyArraySchema and ExternalTorchTensorSchema on the Embeddings page.