Skip to content

Bridge Classification

USGS LiDAR Bridge Point Cloud Classification Pipeline


Overview

An end-to-end pipeline for classifying 3D LiDAR point clouds around bridges from USGS 3DEP data into 4 semantic classes using a sparse 3D U-Net. The pipeline covers data acquisition through inference, with automated weak supervision for scalable training data generation (550K+ bridges).

See Architecture for the classification schema and system design.

Documentation

Document Description
Architecture System design, classification schema, full algorithm details
Data Pipeline Step-by-step data flow with shapes at each stage
AWS Batch Inference Scaling inference with AWS Batch array jobs
Deployment Runbook End-to-end guide: setup, deploy, train, run inference, troubleshoot
API Reference Auto-generated API documentation from source docstrings
Design Decisions Rationale for key architectural choices

Getting Started

See the README for full installation and pipeline run commands.