1 data_handling.efm_views

data_handling.efm_views

Typed, zero-copy views over EFM-formatted ATEK samples.

These classes mirror the style of aria_nbv.data.views but read the keys produced by efm3d.dataset.efm_model_adaptor.load_atek_wds_dataset_as_efm. All properties surface rich shape/type information to make downstream use explicit and safe.

The view layer defines which fields are actor-visible (rgb, calibrated poses, semidense points, EVL/OBB predictions) and which are oracle/evaluation assets (attached GT meshes and GT OBBs). It avoids copying large tensors so dataset writers can attach mesh and snippet context without changing the immutable VIN offline-cache rows.

1.1 Classes

Name Description
EfmGtCameraObbView Per-camera oriented bounding boxes for a single timestamp (EFM schema).
EfmGtTimestampView EFM GT slice for one timestamp across cameras.
EfmGTView Ground-truth annotations (EFM schema) for a snippet.
EfmCameraView Zero-copy camera stream view in EFM schema (images, calibration, timing, optional depth).
EfmTrajectoryView World-frame rig trajectory aligned to snippet frames.
EfmPointsView Padded semi-dense SLAM point cloud view with per-frame metadata.
EfmObbView Snippet-level oriented bounding boxes in the EFM ObbTW layout.
EfmSnippetView Typed wrapper over an EFM-formatted sample plus optional mesh.
VinSnippetView Minimal snippet payload for VIN v2 batching.

1.2 Functions

Name Description
is_efm_snippet_view_instance Return whether value behaves like an EfmSnippetView.
is_vin_snippet_view_instance Return whether value behaves like a VinSnippetView.