Core & Constants¶

Shared constants, enums, and lightweight utilities that form the foundation of the pipeline. These modules have no heavy dependencies (no torch, spconv, pdal, numpy) and can be imported anywhere.

Constants¶

Constant	Value / Type	Description
`NUM_CLASSES`	`4`	Number of model output classes
`CLASS_NAMES`	dict	`{0: "Background", 1: "Ground/Water", 2: "Bridge Deck", 3: "Obstacles"}`
`CLASS_COLORS`	dict	Matplotlib colors per class (`black`, `orange`, `blue`, `yellow`)
`CLASS_COLORS_HEX`	dict	High-contrast hex palette for publication figures
`BRIDGE_DECK_MODEL_CLASS`	`2`	Model class for bridge deck
`BRIDGE_DECK_ASPRS_CODE`	`17`	ASPRS code for bridge deck
`OBSTACLES_MODEL_CLASS`	`3`	Model class for obstacles
`OBSTACLES_ASPRS_CODE`	`18`	ASPRS code for obstacles
`VOXEL_SIZE`	`0.1`	Default voxel size in meters
`SPATIAL_SHAPE_PADDING`	`10`	Padding added to voxel grid spatial shape
`MIN_POINT_COUNT`	`100`	Skip files with fewer points
`BRIDGE_TIMEOUT`	`150`	Default per-bridge timeout in seconds
`AWS_MAX_RETRIES`	`3`	Max S3 retry attempts (adaptive mode)

Class Mapping (ASPRS to Model)¶

Used at preprocessing/training time (LAS_TO_MODEL_MAP):

ASPRS Code	ASPRS Name	Model Class	Model Name
2	Ground	1	Ground/Water
9	Water	1	Ground/Water
17	Bridge Deck	2	Bridge Deck
18	High Noise	3	Obstacles
all others	-	0	Background

Used at inference time (MODEL_TO_LAS_MAP):

Model Class	Model Name	ASPRS Code	ASPRS Name
0	Background	1	Unclassified
1	Ground/Water	2	Ground
2	Bridge Deck	17	Bridge Deck
3	Obstacles	18	High Noise

src.constants ¶

Bridge Classification - shared constants and lightweight utilities.

This module has NO heavy dependencies (no torch, spconv, pdal, numpy) so it can be safely imported anywhere, including environments without a GPU.

InferenceResult ¶

Bases: Enum

Outcome of a single bridge inference call.

InferenceMode ¶

Bases: Enum

Output mode for bridge classification inference.

BridgeTimeout ¶

Bases: BaseException

Raised when a single bridge exceeds the per-bridge wall-clock timeout.

bridge_timeout_guard ¶

bridge_timeout_guard(seconds: float) -> Iterator[None]

SIGALRM-based wall-clock timeout. Raises BridgeTimeout after seconds.

Note: SIGALRM is delivered but the Python handler only fires when the bytecode eval loop runs. This CANNOT interrupt blocking C extensions (e.g. PDAL network reads). For C-blocking code, use subprocess-based timeout with Process.terminate() instead - see _run_bridge_in_subprocess in download_and_weak_supervise_hucs.py.

Parameters:

Name	Type	Description	Default
`seconds`	`float`	Wall-clock timeout in seconds. If <= 0, the guard is a no-op.	required

Raises:

Type	Description
`BridgeTimeout`	If the guarded block exceeds the timeout.

Example

try:
    with bridge_timeout_guard(300):
        result = long_running_operation()
except BridgeTimeout:
    handle_timeout()

src.logging_utils ¶

Shared logging configuration for long-running data scripts.

Provides a file + console logging setup used by download_bridge_lidar and download_and_weak_supervise_hucs (and any future data pipelines).

setup_logging ¶

setup_logging(name: str, log_dir: str = './logs') -> Logger

Set up logging to both file and console.

Creates a timestamped log file under log_dir and returns a logger that writes INFO+ to the file and WARNING+ to stderr.

Parameters:

Name	Type	Description	Default
`name`	`str`	Logger name and log-file prefix (e.g. 'bridge_processing').	required
`log_dir`	`str`	Directory to store log files.	`'./logs'`

Returns:

Type	Description
`Logger`	Configured logger instance.