https://github.com/ajhamdi/SADA
SADA: Semantic Adversarial Diagnostic Attacks for Autonomous Applications (AAAI 2020)
https://github.com/ajhamdi/SADA
aaai2020 adversarial-attacks autonomous-applications blender deep-learning sada self-driving-car
Last synced: 3 months ago
JSON representation
SADA: Semantic Adversarial Diagnostic Attacks for Autonomous Applications (AAAI 2020)
- Host: GitHub
- URL: https://github.com/ajhamdi/SADA
- Owner: ajhamdi
- Created: 2018-07-19T08:57:16.000Z (about 7 years ago)
- Default Branch: master
- Last Pushed: 2021-08-24T05:53:56.000Z (about 4 years ago)
- Last Synced: 2025-04-23T20:17:09.761Z (6 months ago)
- Topics: aaai2020, adversarial-attacks, autonomous-applications, blender, deep-learning, sada, self-driving-car
- Language: Python
- Homepage: https://arxiv.org/abs/1812.02132
- Size: 1.47 MB
- Stars: 12
- Watchers: 2
- Forks: 2
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# SADA: Semantic Adversarial Diagnostic Attacks for Autonomous Applications (AAAI 2020)
### [Paper](https://aaai.org/Papers/AAAI/2020GB/AAAI-HamdiA.1281.pdf) | [video](https://youtu.be/clguL24kVG0) | [supplementary material](https://arxiv.org/pdf/1812.02132.pdf)
Tensorflow implementation of [the paper](https://arxiv.org/abs/1812.02132) in [AAAI 2020](https://aaai.org/Conferences/AAAI-20/). The paper tries to address the robustness of Deep Neeural Networks, but not from pixel-level perturbation lense, rather from semantic lense in which the perturbation happens in the latent parameters that generate the image. This type of robustness is important for safety-critical applications like self-driving cars in which tolerance of error is very low and risk of failure is high.
[SADA: Semantic Adversarial Diagnostic Attacks for Autonomous Applications](https://aaai.org/Papers/AAAI/2020GB/AAAI-HamdiA.1281.pdf)
[Abdullah Hamdi](https://abdullahamdi.com/), [Matthias Muller](https://matthias.pw/), [Bernard Ghanem](http://www.bernardghanem.com/)
## Citation
If you find this useful for your research, please use the following.
```bibtex
@inproceedings{hamdi2020sada,
title = {{SADA:} Semantic Adversarial Diagnostic Attacks for Autonomous Applications},
author = {Abdullah Hamdi and
Matthias Muller and
Bernard Ghanem},
booktitle = {AAAI Conference on Artificial Intelligence},
year = 2020
}
```
## Prerequisites
- Linux
- Python 2 or 3
- NVIDIA GPU (11G memory or larger) + CUDA cuDNN
- [Blender 2.79](https://www.blender.org/download/releases/2-79/)
## Getting Started
### Installation
- install [Blender](https://www.blender.org/download/releases/2-79/) with the version `blender-2.79b-linux-glibc219-x86_64` and add it to your `PATH` by adding the command `export PATH="${PATH}:/home/PATH/TO/blender-2.79b-linux-glibc219-x86_64"` in `/home/.bashrc` file . make sure at the end that you can run `blender` command from your shell script.
- Clone this repo:
```bash
git clone https://github.com/ajhamdi/SADA
cd SADA
```
- install the following `conda` environment as follows:
```bash
conda env create -f environment.yaml
conda activate sada
```
- Download the dataset that contains the 3D shapes and the environments from [this link](https://drive.google.com/drive/folders/1IFKOivjYXBQOhnc2WV7E4hipxCtSoB4u?usp=sharing) and place the folder in the same project dir with name `3d/training_pascal`.
- Download the weights for YOLOv3 from [this link](https://drive.google.com/file/d/1FeHobYulruf98ZOnWVpmqK8vza2u6MX-/view?usp=sharing) and place in the `detectos` dir.
### Dataset
- We collect 100 3D shapes from 10 classes from [ShapeNet](https://www.shapenet.org/) and Pascal3D . All the sahpes are available inside the blender environment `3d/training_pascal/training.blend` file. The classes are the following
1. **aeroplane**
1. **bench**
1. **bicycle**
1. **boat**
1. **bottle**
1. **bus**
1. **car**
1. **chair**
1. **dining table**
1. **motorbike**
1. **train**
1. **truck**
- The parameters that control the environment are 8 as follows
1. **camera distance to the object**
1. **camera azimuth angle**
1. **camera pitch angle**
1. **light source azimuth angle**
1. **light source pitch angle**
1. **color of the object (R-channel)**
1. **color of the object (G-channel)**
1. **color of the object (B-channel)**
## Generating images from the 3D environment for a specific class with random parameters and storing the 2D dataset in the folder `generated`
```
python main.py --is_gendist=True --class_nb= 0 --dataset_nb= 0 --gendist_size= 10000
```
* `is_gendist` : is the option to generate distribution of parameters and images
* `class_nb` the class of the 12 classes above to generate
* `dataset_nb` : is the number assigned to the dataset generated
* `gendist_size` : the number of inages generated
## training BBGAN
```
python main.py --is_train=True --valid_size=50 --log_frq=10 --batch_size=32 --induced_size=50 --nb_steps=600 --learning_rate_t=0.0001 --learning_rate_g=0.0001
```
* `class_nb` the class of the 12 classes above to generate
* `dataset_nb` : is the number assigned to the dataset generated
* `nb_steps` : is the number of training steps of the GAN
* `log_frq=10` : how often u save the weights of the network
* `induced_size`: is the number of best samples that will be picked out of the total numberof generated images
* `learning_rate_g`: the learning rate forf the generator
* `learning_rate_t`: the learning rate forf the discrminator
* `valid_size` : is the number of paramters u will be generating eventually for evaluation of the BBGAN
## Self-Driving with [CARLA](http://carla.org/)
* coming soon
## UAV racing with [Sim4CV](https://sim4cv.org/)
* coming soon