SRLT: Scanning Radar Labelling Tool for annotating segmentation mask

SRLT is a graphical scanning radar image annotation tool.

It's written in Python and uses Qt for its graphical interface.

This tool currently supports the Oxford Radar Robotcar Dataset, you can also modified it to use in custom dataset.

Description

Features

Radar image segmentation annotation.
Segment Anything Model(SAM) supported semi-auto annotation

Folder Structure

SRLT/
├── data/
│   ├── prepare_data.py
│   ├── gen_lidar_mask.py
│   └── mask2box.py
├── gui/
│   ├── utils/
│   │   ├── file_utils.py
│   │   ├── gui_utils.py
│   │   └── image_method.py
│   ├── interact/
│   │   ├── interaction.py
│   │   └── timer.py
│   ├── btn_controller.py
│   ├── controller.py
│   ├── inference_core.py
│   ├── shortcut.py
│   └── ui.py
├── models
├── itri.sh
├── oxford.sh
├── setup.py
└── start.py

Getting Started

Prerequisites

1. Install Ian's toolbox library

This is the custom library that contains multiple useful functions to deal with radar image , LiDAR publish, and transforms.. etc

pip install iantoolbox

2. Install the Segment Anything Model(SAM)

Install the SAM for semi-auto annotatation

pip install git+https://github.com/facebookresearch/segment-anything.git

2. Download the necessary files

For example : one sequence among Oxford Radar RobotCar Dataset.

In this annotation tool, the necessary files include:

Navtech CTS350-X Radar - raw radar data
Navtech CTS350-X Radar Optimised SE2 Odometry -used to compensate the ego-motion
(Optional) Velodyne HDL-32E Left Pointcloud- (used for lidar BEV reference)
(Optional) Velodyne HDL-32E Right Pointcloud- (used for lidar BEV reference)

3. Preprocess the required data

Generate the compensated $N$ radar scans (Current + $N!-!1$ historical scans)

/data/prepare_data.py

(Optional) Generate the LiDAR BEV images for reference

/data/gen_lidar_mask.py

The final folder structure for data will looks like

data/
├── radar/
│   ├── xxxx.png
│   └── ...
├── radar_history/
│   ├── xxxx_1.png
│   ├── xxxx_2.png
│   └── ...
└── lidar_mask/
    └── xxxx.png

Installation

Clone the repository

git clone https://github.com/Ianpengg/SRLT

Setup the environment

# Setup conda
conda create -n srlt python=3.7.5
conda activate srlt

# Install requirement
pip install -r requirements.txt

Usage

Open the GUI

oxford.sh

Reference

Parts of the code have been adapted from the following repository: https://github.com/ori-mrg/robotcar-dataset-sdk

LICENSE

This project is licensed under the terms of the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. For the full license, please refer to the LICENSE file.