notesum.ai
Published at November 27Visual Adversarial Attack on Vision-Language Models for Autonomous Driving
cs.CV
Released Date: November 27, 2024
Authors: Tianyuan Zhang1, Lu Wang1, Xinwei Zhang1, Yitong Zhang1, Boyi Jia1, Siyuan Liang2, Shengshan Hu3, Qiang Fu1, Aishan Liu1, Xianglong Liu1
Aff.: 1Beihang University, China; 2National University of Singapore, Singapore; 3Huazhong University of Science and Technology, China

| Method/Metrics | Weather | Traffic. | Time | Scene | Open. | Desc | Final |
|---|---|---|---|---|---|---|---|
| Raw | 48.75 | 52.82 | 39.71 | 43.31 | 29.79 | 41.37 | 42.63 |
| FGSM [9] | 47.27 | 52.99 | 46.62 | 45.46 | 23.60 | 45.71 | 43.61 |
| PGD [39] | 32.43 | 40.67 | 36.51 | 33.04 | 22.15 | 36.83 | 33.60 |
| AttackVLM [66] | 44.32 | 50.60 | 41.12 | 43.25 | 28.98 | 42.51 | 41.96 |
| AnyAttack [61] | 45.10 | 51.12 | 43.28 | 44.10 | 27.25 | 43.14 | 42.28 |
| ADvLM | 31.09 | 41.06 | 39.36 | 32.60 | 17.44 | 36.45 | 32.99 |