pasture

A Rust library for working with point cloud data. It features:

• Fine-grained support for arbitrary point attributes, similar to PDAL, but with added type safety
• A very flexible memory model, natively supporting both Array-of-Structs (AoS) and Struct-of-Arrays (SoA) memory layouts (which pasture calls 'interleaved' and 'columnar')
• Support for reading and writing various point cloud formats with the pasture-io crate (such as LAS, LAZ, 3D Tiles, as well as ASCII files)
• A growing set of algorithms with the pasture-algorithms crate

To this end, pasture chooses flexibility over simplicity. If you are looking for something small and simple, for example to work with LAS files, try a crate like las. If you are planning to implement high-performance tools and services that will work with very large point cloud data, pasture is what you are looking for!

Usage

Add this to your Cargo.toml:

[dependencies]
pasture-core = "0.4.0"
# You probably also want I/O support
pasture-io = "0.4.0"

Here is an example on how to load a pointcloud from an LAS file and do something with it:

use anyhow::{bail, Context, Result};
use pasture_core::{
    containers::{BorrowedBuffer, VectorBuffer},
    layout::attributes::POSITION_3D,
    nalgebra::Vector3,
};
use pasture_io::base::{read_all};

fn main() -> Result<()> {
    // Reading a point cloud file is as simple as calling `read_all`
    let points = read_all::<VectorBuffer, _>("pointcloud.las").context("Failed to read points")?;

    if points.point_layout().has_attribute(&POSITION_3D) {
        for position in points
            .view_attribute::<Vector3<f64>>(&POSITION_3D)
            .into_iter()
            .take(10)
        {
            println!("({};{};{})", position.x, position.y, position.z);
        }
    } else {
        bail!("Point cloud files has no positions!");
    }

    Ok(())
}

For more examples, check out the pasture_core examples and the pasture_io examples.

Migration from versions < 0.4

With version 0.4, the buffer API of pasture-core was rewritten. If you are migrating from an earlier version, here are some guidelines for migration. Also check out the documentation of the containers module.

New buffer types

The main buffer types were renamed:

• InterleavedVecPointStorage is now VectorBuffer
• PerAttributeVecPointStorage is now HashMapBuffer

The trait structure is also different:

• PointBuffer and PointBufferWriteable are replaced by BorrowedBuffer, BorrowedMutBuffer, and OwningBuffer, which define the ownership model of the buffer memory
• InterleavedPointBuffer and InterleavedPointBufferMut are now InterleavedBuffer and InterleavedBufferMut
• PerAttributePointBuffer and PerAttributePointBufferMut are now ColumnarBuffer and ColumnarBufferMut. In general, the term PerAttribute is replaced by the more common term Columnar

There are no more extension traits (e.g. PointBufferExt). To get/set strongly typed point data, you now use views which can be obtained through the BorrowedBuffer and BorrowedBufferMut traits:

let view = buffer.view_attribute::<Vector3<f64>>(&POSITION_3D);

Views support strongly typed access to the data and are convertible to iterators.

New interface for readers and writers

The PointReader and PointWriter traits are no longer object safe. Instead, they have read and read_into methods that are strongly typed over the buffer type for improved efficiency. There is a GenericPointReader type, which uses static dispatch and encapsulates readers for LAS, LAZ, and 3D Tiles.

Development

pasture is in the early stages of development and bugs may occur.

License

pasture is distributed under the terms of the Apache License (Version 2.0). See LICENSE for details.