Adversarial-Traffic-Signs

Supplementary Material: Click this Dropbox link to download. It contains 3 files and you can place them into corresponding directories to run the code.

Contents

• 1.CTSRD_Classification: Train a Chinese traffic sign classification model using CTSRD Dataset. It uses ResNet101 in Keras.
• 2.DARTS-Attack-on-CTSRD: Modified DARTS pipeline, it can generate adversarial Chinese traffic signs.
• 3.TT100K_ObjectDetection: Using an object detection model pre-trained by TT100K Dataset to do test of adversarial traffic signs in real world. It uses Faster R-CNN-ResNet101 in TensorFlow Object Detection API.
• 4.Examples of Adversarial Traffic Signs: Some examples.

Acknowledgement

This project is what I do in my thesis of B.Eng. in Electronic and Information Engineering.
I appreciate support from VeriMake, issues are also welcome on VeriMake BBS (in Chinese).

Reference

• Chinese Traffic Sign Database
• Keras API Reference
• DARTS: Deceiving Autonomous Cars with Toxic Signs
• TensorFlow Guide
• Tsinghua-Tencent 100K Dataset
• TensorFlow Object Detection API