Home

Awesome

SwiftFusion

A differentiable sensor fusion library written in Swift.

Think factor graphs (à la GTSAM) coupled with deep learning (via Swift for TensorFlow).

This project is in an early phase, but feel free to explore! Subject to massive changes. :-)

Getting started

Installation

macOS

Download a Swift for TensorFlow macOS toolchain and follow the macOS installation instructions. Installing the latest development snapshot is recommended. Check here for the "required" version of Xcode for using nightly toolchains. Sometimes toolchains may also work with older Xcode minor versions than the one listed.

To use Swift for TensorFlow, add the toolchain to PATH:

export PATH="/Library/Developer/Toolchains/swift-tensorflow-RELEASE-0.12.xctoolchain/usr/bin/:$PATH"

Linux

Download a Swift for TensorFlow Linux toolchain and follow the Linux installation instructions. Installing the latest development snapshot is recommended.

NVIDIA Jetson

Download the Swift for TensorFlow NVIDIA Jetson toolchain and follow the Linux installation instructions.

Testing

swift test --enable-test-discovery

Benchmarking

swift run -c release -Xswiftc -cross-module-optimization SwiftFusionBenchmarks

Development

Xcode (macOS)

Xcode provides a great Swift development experience on macOS.

To open SwiftFusion in Xcode, run open Package.swift.

Visual Studio Code

To enable Swift autocomplete in Visual Studio Code, install the Maintained Swift Development Environment plugin, and set the following plugin settings:

"sde.languageServerMode": "sourcekit-lsp",
"sourcekit-lsp.serverPath": "<your toolchain path>/usr/bin/sourcekit-lsp",
"sourcekit-lsp.toolchainPath": "<your toolchain path>",
"swift.path.swift_driver_bin": "<your toolchain path>/usr/bin/swift",

The CodeLLDB plugin enables debugging within Visual Studio Code. You need to set the following setting:

"lldb.library": "/swift-tensorflow-toolchain/usr/lib/liblldb.so"

A sample launch.json file:

{
  "version": "0.2.0",
  "configurations": [
    {
      "type": "lldb",
      "request": "launch",
      "name": "Debug",
      "program": "${workspaceFolder}/.build/x86_64-unknown-linux-gnu/debug/SwiftFusionPackageTests.xctest",
      "args": [
        "--enable-test-discovery"
      ],
      "cwd": "${workspaceFolder}"
    }
  ]
}

Code overview

The main code is in Sources/SwiftFusion, organized under subdirectories.

Core

The core abstractions of SwiftFusion:

Specific Vector-like data structures:

Additional utilities:

Inference

A factor graph stores factors in a storage array of type [ObjectIdentifier: AnyFactorArrayBuffer]. ObjectIdentifier is a built-in Swift identifier type and AnyFactorArrayBuffer is defined in FactorsStorage.swift.

Optimizers

MCMC

Geometry

This directory defines rotation and pose types.

We use the GTSAM naming scheme: Pose2, Rot2, Pose3, and Rot3.

2D and 3D points are simply represented as Vector2 and Vector3.

Image

Datasets

This directory primarily contains support for reading G2O files.

LICENSE

Copyright 2020 The SwiftFusion Team

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.