vision

Low-level calibration and pointcloud foundation libraries

Quick Start

Install

sudo bash scripts/deps/install_bazel.sh

Build

bazel build //calibration:all

Test

bazel test //calibration:all

GPU Lidar Build

For the GPU-native lidar package:

bazel test --config=gpu --@rules_cuda//cuda:archs=compute_89:sm_89 //pointcloud:all

Or build in the provided CUDA container:

docker/scripts/whl.sh 'bazel test //pointcloud:all --config=gpu --@rules_cuda//cuda:archs=compute_89:sm_89'

Public C++ API

Camera

Canonical dependency and include:

#include "calibration/camera_model.h"

@vision//calibration:calibration

The camera module provides self-implemented projection and back-projection models:

• PinholeCameraModel
• BrownCameraModel with OpenCV-compatible (k1, k2, p1, p2, k3) ordering
• OmnidirectionalCameraModel with OpenCV omnidir-compatible xi and (k1, k2, p1, p2) ordering

Lidar

Canonical dependency and include:

#include "pointcloud/core/point_cloud.h"

@vision//pointcloud:pointcloud

The first lidar release exposes:

• point cloud data interfaces with planar/interleaved layout support
• unified/device buffer ownership and fixed-size buffer pooling
• stream-aware GPU-native spherical Range Image generation
• GPU DecimationFilter
• GPU CropBoxFilter for finite-value, box, and height-window style filtering
• a path toward DriveWorks-class lidar base-library parity documented under docs/context/

Context Docs

Project context, references, and engineering TODOs live under:

docs/context/