Publications

Created a diagnostic evaluation framework for AI-generated driving videos: a failure modes taxonomy, a AI driving videos benchmark, and a driving-aware evaluator combining static/temporal cues.

AIGVE-Tool introduces a unified, modular framework, a benchmark (AIGVE-Bench) for comprehensive, standardized evaluation of AI-generated videos across multiple quality dimensions, and a structured five-category taxonomy of evaluation metrics.

A novel open-vocabulary multi-modal 3D object detection method which can incorporate multi-view input and detect complex events (including attributes) without needing novel class anchor size; proposed a dataset for open-vocabulary attribute detection

A comprehensive review of current methodologies of AI-Generated Video Evaluation (AIGVE), highlighting emerging field, and guiding for future research directions.

A general framework with efficient feature extractions and layers (Sparse Down-Sampling, Global Spatial Aggregation). It achieves great accuracy-speed performance with relatively low memory cost under several perception tasks (including 3D/2D detection/segmentation).

A novel 3D Object Detection method that structure the task as a denoising diffusion process from noisy 3D boxes to target boxes.

A ViT-pured hierarchical architecture that embed both Images and Point Clouds, fused via a mixViT for efficient 3D Object Detection.

A BSO-based collaborative searching framework for swarm robotics.

An multitarget searching method for robotic swarms.