Reference

This page provides a guide to the BigQuery tables used for training models, including their schemas and field descriptions. You'll also find a breakdown of the core concepts—snapshots, assets, and observations—that structure the data, along with details on how to use the tables where relevant.

snapshots

A snapshot is a fixed, unchangeable copy of a dataset at a specific moment in time. The snapshot table provides metadata associated with snapshots, to allow you to understand the temporal state of the data at a high level.

Column name	Type	Description
snapshot_id	STRING	Unique identifier for the snapshot. Used as a key to join tables.
subscription_id	STRING	Unique identifier for the subscription.
creation_time	TIMESTAMP	ISO 8601 formatted timestamp e.g., 2019-09-25 17:26:27.757171 UTC .

all_observations

An "observation" is a sighting of an asset in the real world. Observations differ from assets in that they contain metadata for the detection of an asset, i.e., the time the image was captured and the position of the camera.

The all_observations table provides observations from all snapshots. You can use this table to detect differences in observations between snapshots.

Column name	Type	Description
snapshot_id	STRING	Unique identifier for the snapshot. Used as a key to join tables.
asset_id	STRING	Unique identifier for the asset.
asset_type	STRING	Major classification of the asset e.g., ASSET_CLASS_ROAD_SIGN .
location	STRUCT	Struct containing lat/lng coordinates as floats.
detection_time	TIMESTAMP	ISO 8601 formatted timestamp e.g., 2019-09-25 17:26:27.757171 UTC .
observation_id	STRING	String that uniquely identifies the observation.
bbox	STRUCT	Struct of structs containing the x/y coordinates aligning the asset's bounding box.
camera_pose	STRUCT	Struct containing floats for lat/lng, altitude (in meters), pitch, heading, and roll.
capture_time	TIMESTAMP	ISO 8601 formatted timestamp e.g., 2019-09-25 17:26:27.757171 UTC .
gcs_uri	STRING	Google Cloud Storage URI where the image is hosted.
map_url	STRING	Google Maps URL that shows the location of the observation.

all_assets

An "asset" is an object in the real world. The all_assets table provides assets from all snapshots. You can use this table to detect differences in assets between snapshots.

Column name	Type	Description
asset_id	STRING	Unique identifier for the asset.
snapshot_id	STRING	Unique identifier for the snapshot. Used as a key to join tables.
asset_type	STRING	Major classification of the asset e.g., ASSET_CLASS_ROAD_SIGN .
observation_id	STRING	String that uniquely identifies the observation.
location	STRUCT	Struct containing lat/lng coordinates as floats.
detection_time	TIMESTAMP	ISO 8601 formatted timestamp e.g., 2019-09-25 17:26:27.757171 UTC .

latest_observations

The latest_observations table provides observations only from the most recent snapshot.

Column name	Type	Description
snapshot_id	STRING	Unique identifier for the snapshot. Used as a key to join tables.
asset_id	STRING	Unique identifier for the asset.
asset_type	STRING	Major classification of the asset e.g., ASSET_CLASS_ROAD_SIGN .
location	STRUCT	Struct containing lat/lng coordinates as floats.
detection_time	TIMESTAMP	ISO 8601 formatted timestamp e.g., 2019-09-25 17:26:27.757171 UTC .
observation_id	STRING	String that uniquely identifies the observation.
bbox	STRUCT	Struct of structs containing the x/y coordinates aligning the asset's bounding box.
camera_pose	STRUCT	Struct containing floats for lat/lng, altitude (in meters), pitch, heading, and roll.
capture_time	TIMESTAMP	ISO 8601 formatted timestamp e.g., 2019-09-25 17:26:27.757171 UTC .
gcs_uri	STRING	Google Cloud Storage URI where the image is hosted.
map_url	STRING	Google Maps URL that shows the location of the observation.

latest_assets

The latest_assets table provides assets only from the most recent snapshot.

Column name	Type	Description
asset_id	STRING	Unique identifier for the asset.
snapshot_id	STRING	Unique identifier for the snapshot. Used as a key to join tables.
asset_type	STRING	Major classification of the asset e.g., ASSET_CLASS_ROAD_SIGN .
observation_id	STRING	String that uniquely identifies the observation.
location	STRUCT	Struct containing lat/lng coordinates as floats.
detection_time	TIMESTAMP	ISO 8601 formatted timestamp e.g., 2019-09-25 17:26:27.757171 UTC .