Stray Studio - 3D Data Annotation Tool

This project is a graphical user interface for annotating point clouds and Stray scenes.

Several different label types are supported, including:

• Bounding boxes
• Keypoints
• Rectangles

Bounding Box Labels

An example of a bounding box is shown above. They have the following properties:

• class_id the class id of the label.
• position the x, y, z position relative to the point cloud.
• orientation the rotation that transforms vectors in the box coordinate frame to the point cloud coordinate frame.
• dimensions three values denoting the width along the x, y and z axes in the bounding boxes local coordinate frame.

Keypoint Labels

Keypoints are individual points in the global coordinate frame. They have the following properties:

• class_id the class id of the label.
• position the x, y, z position in the global frame.

Rectangle Labels

Rectangles, show above, are rectangular planes that have a size (height and width), an orientation and position.

The properties are:

• class_id the class id of the label.
• position the x, y, z position of the center in the global frame.
• orientation the rotation taking vectors in local frame to the world frame.
• size width and height of the rectangle.

Installation

Install Dependencies - Mac

1. Install Homebrew: https://brew.sh/
2. Run brew update && brew install cmake libomp eigen boost git-lfs pkg-config

Install Dependencies - Linux

1. RUN sudo apt install libeigen3-dev libglfw3-dev libomp-dev libxinerama-dev libxcursor-dev libxi-dev git-lfs cmake libboost-all-dev

Build and install the annotation tool

1. Get the source code git clone https://github.com/StrayRobots/3d-annotation-tool.git (requires git)
2. Navigate to the 3d-annotation-tool directory (e.g. cd 3d-annotation-tool)
3. Initialize git submodules with git submodule update --init --recursive
4. Create a build directory with mkdir build
5. Pull git-lfs objects (helper meshes etc) with git lfs install && git lfs pull
6. To build the project run cd build && cmake .. && make -j8 (-j8 specifies the number of parallel jobs, for a fewer jobs use a lower number (than 8))
7. The executable is called studio, it can be executed with ./studio <path-to-pointcloud>. You can find an example cloud.ply point cloud from here. <path-to-pointcloud> should then specify the absolute path to the downloaded file.

Usage

Run ./studio <path-to-pointcloud> to open a point cloud in the viewer. Currently only .ply point clouds are supported. Annotations are saved into a file of with the same filename but a .json file extension. When annotating point clouds, you can move to the next point cloud in the same directory as <path-to-pointcloud> using tab.

An example cloud.ply point cloud can be downloaded from here.

The keyboard shortcuts are:

• ctrl+s to save the annotations.
• k switches to the keypoint tool.
• b switches to the bounding box tool.
• r switches to the rectangle tool.
• v switches to the move tool.

The Stray Toolkit

This project is part of the Stray command line interface, a toolkit to make building 3D computer vision applications easy. Stray Studio can be used through the stray studio command.

The Stray toolkit allows you to build point clouds out of handheld or robot mounted RGB-D scans. It also contains utility functions to manage datasets and export labels into commonly used formats, such as Yolo for object detection.

Running tests

In your build directory, run:

cmake .. -DBUILD_TESTS=1
make build_tests
ctest

Code formatting

Code can be formatted using clang-format.

Install clang-format on Mac (brew):

brew install clang-format

Install clang-format on Ubuntu:

sudo apt install clang-format

There's a script that runs formatting on all .h and .cc files based on the .clang-format file. Usage: ./run_formatting