Targeted-Attack-on-Handcrafted-Features

We propose a novel targeted adversarial attack for multiple well-known handcrafted pipelines and datasets. Our attack is able to match an image with any given target image which can be completely different from the original image. Our approach provides a trade-off between effectiveness and imperceptibility, and outperforms the baselines on both metrics

Can you guess the best matching image?

Guess.Game.mp4

Selective Average Explained

selective.average.mp4

Decision Based Attack Explained

Decision.based.attack.mp4

Comparison with the Baselines

comparison.mp4

Showed attack on Place Recognition: FABMAP

FABMAP.Attack.mp4

Showed attack on 3D Reconstruction: COLMAP

colmap.attack.mp4

Showed Attack on ORB_SLAM3 Tracking

attack.tracking.mp4

Requirments

• adversarial-robustness-toolbox 1.9.1
• opencv-contrib-python 3.4.2.17
• PCV
• torch
• visvis
• BOW Matlab Code

Formatting

1. Folder1: Base Images
   • fallbase.jpg
   • springbase.jpg
   • summerbase.jpg
   • winterbase.jpg
2. Folder2: Target Classes
   • Fall
     • 1.jpg
     • 2.jpg
     • ...
   • Spring
     • 1.jpg
     • 2.jpg
     • ...
   • Summer
     • 1.jpg
     • 2.jpg
     • ...
   • Winter
     • 1.jpg
     • 2.jpg
     • ...
3. Folder3: Save Folder
   • Fall
     • 1.jpg
     • 2.jpg
     • ...
   • Spring
     • 1.jpg
     • 2.jpg
     • ...
   • Summer
     • 1.jpg
     • 2.jpg
     • ...
   • Winter
     • 1.jpg
     • 2.jpg
     • ...