NeurIPS 2026
Relational Feature Distillation for Lightweight 3D Point Cloud Segmentation was accepted to NeurIPS 2026.
Publication
Amir Salarpour
Postdoctoral Research Associate
School of Computing, Clemson University · TigerSec Lab
I work on secure and trustworthy perception for autonomous systems, with emphasis on adversarial machine learning, vision-language models, and efficient 3D vision.
Recent news
ACSAC 2026
On the Vulnerability of Lane Detectors to Physical Shadow Attack was accepted to ACSAC 2026.
Publication
IROS 2026
Distilling Vision-Language Models for Robust Traffic Sign Perception in Autonomous Vehicles was accepted to IROS 2026.
Publication
Research focus
Secure Autonomous Perception
Physical and black-box attacks, realistic evaluation, and defenses for camera- and LiDAR-based perception.
Vision-Language Models for Driving
Robust multimodal reasoning, traffic-scene understanding, and test-time adaptation for driving systems.
3D Vision and Point Clouds
Efficient methods for point-cloud recognition, segmentation, tracking, and geometric representation learning.
Trustworthy Machine Learning
Reliability under distribution shift, uncertainty-aware deployment, and interpretable failure analysis.
Selected publications
Meta-VLM: Which Metadata Do Vision-Language Models Need to Drive Safely?
IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2027.
Relational Feature Distillation for Lightweight 3D Point Cloud Segmentation
Conference on Neural Information Processing Systems (NeurIPS), 2026.
On the Vulnerability of Lane Detectors to Physical Shadow Attack
IEEE Annual Computer Security Applications Conference (ACSAC), 2026.
GATE: Reliability-Gated Gaussian Evidence Fusion for Training-Free Test-Time Adaptation of Vision-Language Models
British Machine Vision Conference (BMVC), 2026.
Distilling Vision-Language Models for Robust Traffic Sign Perception in Autonomous Vehicles
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2026.
Budget-Aware Adaptive Adversarial Patches for Black-Box Object Detection
IEEE International Conference on Image Processing (ICIP), 2026.
Selected projects
Active
Resilient Autonomous Vehicle Perception Under Adversarial Settings
Research on realistic physical attacks and defenses for camera- and LiDAR-based autonomous-vehicle perception, with an emphasis on robustness under changing viewpoints, distance, and environmental conditions.
autonomous driving · adversarial perception · camera and LiDAR
- Period
- 2025–2026
- Sponsor
- U.S. DOT UTC TraCR
- Role
- Key Personnel