We are the Trustworthy Visual Intelligence Group. We focus on pushing the boundaries of Computer Vision, Embodied Intelligence, and Reliable AI.
The Trustworthy Visual Intelligence (TVI) group is a core part of the Image Cognition Group at Chongqing University of Posts and Telecommunications. Founded in 2020, we are a dynamic research team led by Associate Professor Jiaxu Leng .
Our Focus:
Trustworthy AI
Video Anomaly Detection
Video Analysis
Embodied Intelligence
Our members have led projects including the National Natural Science Foundation of China and the Chongqing Bo Xin Plan. We actively publish in top-tier venues:
The team's latest news, events, and announcements will be continuously updated.
One research work titled “Difficulty-Conditioned Attribute-Specific Restoration for Low-Light Image Enhancement” from the HXJM team was accepted by ECCV 2026. The work proposes DCASR, a difficulty-aware and attribute-specific restoration framework for adaptive low-light image enhancement.
Read more →
Recently, the team’s research work “ZeroTeach: Cross-Embodiment Manipulation from Human Videos via Physics-Informed Diffusion” was accepted by IEEE SMC. The work proposes ZeroTeach, a framework that learns transferable robot manipulation policies directly from human RGB-D videos.
Read more →
Recently, the team’s research work “Causal Bootstrapped Alignment for Unsupervised Video-Based Visible–Infrared Person Re-Identification” was accepted by IEEE Transactions on Information Forensics and Security. The work proposes CBA, an unsupervised framework for learning robust cross-modal video person re-identification representations from unlabeled visible and infrared video tracklets.
Read more →The team's latest news, events, and announcements will be continuously updated.
Garden adopts a collaborative architecture of 'small perception model + large multimodal model'. It enables precise garbage recognition, road cleanliness evaluation, and intelligent scheduling across multiple scenarios in complex terrains.
The team's latest news, events, and announcements will be continuously updated.
The 'Fiery Eyes and Golden Sight' system adopts an integrated architecture of 'human-inspired rule-based systems + large foundation models'. It enables cross-camera person re-identification, face super-resolution, and intelligent text-to-video retrieval, achieving a complete loop from passive monitoring to active analysis.