ECCV 2024Self-Adapting Large Visual-Language Models to Edge Devices across Visual Modalities

Kaiwen Cai, Zhekai Duan, Gaowen Liu, Charles Fleming, Chris Xiaoxuan Lu

   

Unleash the power of large language models for non-RGB modality on edge devices.

AAAI 2025Risk Controlled Image Retrieval

Kaiwen Cai, Chris Xiaoxuan Lu, Xingyu Zhao, Wei Huang and Xiaowei Huang

We make the uncertainty of image retrieval certain.

EWSN 2023Robust Human Detection under Visual Degradation via Thermal and mmWave Radar Fusion

Kaiwen Cai and Qiyue Xia, Peize Li, John Stankovic, Chris Xiaoxuan Lu

  

The majority of human detection methods rely on the sensor using visible lights (e.g., RGB cameras) but such sensors are limited in scenarios with degraded vision conditions. We build a thermal image and mmWave radar-based human detection pipeline, which consists of a Bayesian feature extractor and a novel uncertainty-guided fusion module.

RA-LUncertainty Estimation for 3D Dense Prediction via Cross-Point Embeddings

Kaiwen Cai, Chris Xiaoxuan Lu, Xiaowei Huang

  

We propose a generic and effective framework, CUE, for dense uncertainty estimation of 3D point clouds. And we demonstrate it in two different tasks: (1) in 3D geometric feature learning we for the first time obtain well-calibrated dense uncertainty, and (2) in semantic segmentation we reduce uncertainty’s Expected Calibration Error of the state-of-the-arts by 16.5%. All uncertainties are estimated without compromising predictive performance.

ICRA 2022AutoPlace: Robust Place Recognition with Single-Chip Automotive Radar

Kaiwen Cai, Bing Wang, Chris Xiaoxuan Lu in the 2022 IEEE International Conference on Robotics and Automation (ICRA).

   

We propose a place recognition pipeline for single-chip automotive radar, making the use of Doppler velocity measurement, RCS measurement and a spatial-temporal nerural network.