smart-annotation-pointrcnn

SANE-PointRCNN, a browser-based 3D bounding boxes annotation tool assisted by PointRCNN. This works is based on H. A. Arief's paper and code

Environment

Tested on Debian 9.9, Cuda: 10.0, Python: 3.6, Pytorch: 1.2.0 with Anaconda

Installation

git clone --recursive https://github.com/ziliHarvey/smart-annotation-pointrcnn.git
cd app/PointCNN/
sh build_and_install.sh

Also install all necessary libraries using conda, such as flask, easydict,tqdm, tensorboardX, etc.

Usage

cd app
python app.py

Open your browser and then go to http://0.0.0.0:7772. The first time loading will be relatively slow and the rest will be very fast.
For detailed instructions on annotating your own data, please refer to Docs.

Progress

• Reorganized the code base by running PointRCNN as backend
• Fully-Automated-Bbox click
• Segmented object points display
• One-click annotation by holding A key and click on the point
• Fix heading angle in boxes display
• Display all LiDAR points with corresponded point labels
• Modify dataLoader to run on the specied file
• Modify dataLoader to run for inference without ground truth
• JSON and KITTI-format conversion and offline visualization
• remove legacy code and files and clean
• Tracking, etc.

Contact

Zi Li
Kartik Sah