1 vin.feature_bank.compression

vin.feature_bank.compression

Descriptor compression helpers for point-feature banks.

This module provides explicit channel slicing or linear projection and derives human-readable compression labels for actor-visible descriptor payloads.

1.1 Functions

Name	Description
compress_point_features	Apply an explicit descriptor compression transform.
resolve_compression_id	Return a human-readable label for raw, sliced, or projected descriptors.

1.1.1 compress_point_features

vin.feature_bank.compression.compress_point_features(
    features,
    *,
    projection=None,
    output_dim=None,
)

Apply an explicit descriptor compression transform.

1.1.1.1 Parameters

Name	Type	Description	Default
features	Tensor	Tensor[..., C] point descriptors.	required
projection	Tensor | None	Optional Tensor["C D"] projection matrix.	None
output_dim	int | None	Optional leading channel count when using a simple slice.	None

1.1.1.2 Returns

Name	Type	Description
	Tensor	Compressed descriptors.

1.1.2 resolve_compression_id

vin.feature_bank.compression.resolve_compression_id(
    compression_id,
    *,
    projection,
    output_dim,
    output_channels,
)

Return a human-readable label for raw, sliced, or projected descriptors.

The generated label records only the transform family and output width; it is not a content hash of a projection matrix and cannot uniquely identify learned compression weights.