DANGER: A Danger Awareness Neural Generative Extension Framework for Robustness test

We have released the DANGER Dataset publicly to aid the research community in making advancements in model robustness and explainability.

The DANGER framework is universal to many opensource dataset. In this repo we provided two datasets - the vKITTI1 add-on dataset and the vKITTI2 add-on dataset.

Overview

• Changelog
• Model Zoo
• Tutorials
• Citation

Changelog

[2021-06-09] DANGER v0.1.0 is released.

Introduction

What does DANGER do?

0002 Cut-in Opposite

0018 Cut-in

0006 Lane Change

DANGER design pattern

Model Zoo

vKITTI1

Prerequisites

• Linux
• Python 3.6+
• PyTorch 0.4
• NVIDIA GPU (GPU memory > 8GB) + CUDA 9.0
• Conda

Use our notebook hosted on Colab/Colab Pro will help you setup all the prerequisites automatically.

Sun Jun  5 01:15:01 2022       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 460.32.03    Driver Version: 460.32.03    CUDA Version: 11.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  Tesla P100-PCIE...  Off  | 00000000:00:04.0 Off |                    0 |
| N/A   37C    P0    28W / 250W |      0MiB / 16280MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+
Python 3.6.10 :: Anaconda, Inc.
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2017 NVIDIA Corporation
Built on Fri_Sep__1_21:08:03_CDT_2017
Cuda compilation tools, release 9.0, V9.0.176
time: 2min 19s (started: 2022-06-05 01:12:42 +00:00)

Tutorials

We host our tutorial notebooks on Colab, and we suggest you initialize your vm with a GPU when you setup the 3D-SDN environment.

sudo chmod 755 ./DANGER/install.sh
./DANGER/install.sh

1 3D-SDN environment

3D-SDN environment:

• There are sections in this notebook: Hardware Setup, 3D-SDN Example, Semantic Training, Geometry Training, Textural Training, and test.json
• Please run Hardware Setup to modify the default supported drivers that Colab provided.
• Run 3D-SDN Example to verify the installation, which would help you finish the Getting Started section described in 3D-SDN.
• Use test.json, provided in step 2, to generate your dataset. You can ignore the Semantic Training, Geometry Training, Textural Training steps, but it would be helpful when you interested to train you own model.

2 JSON generator

Generated JSON files are preserved at JSONs.

License

DANGER is released under the MIT License

Acknowledgement

DANGER is an open source project for dataset manipulation and model robustness test. We would like to thank the authors of 3D-SDN for their open-source release.

Citation

If you find this project useful in your research, please consider citing:

@inproceedings{xu2022framework,
  title={A Framework for Generating Dangerous Scenes for Testing Robustness},
  author={Xu, Shengjie and Mi, Lan and Gilpin, Leilani H},
  booktitle={Progress and Challenges in Building Trustworthy Embodied AI},
  year={2022}
}

GitHub Push Code

git add .

git commit -m 'init'

git branch -M main

git push -u origin main

To Do

• Calculate u,v
• Verify u,v by check each operation in vkitti_edit_benchmark.json
• Try generate *.json by generate delete operation
• Try generate *.json by generate ry operation
• Try generate *.json by change u,v operation