NeurIPS 2026 OralCorresponding author
Language-Conditioned World Modeling for Visual Navigation
Connects language instructions, imagined future observations, and action selection in a world model for visual navigation.
Publications
Representative contributions in world models, visual perception, multimodal reasoning, and efficient learning.
World models & visual navigation
From human-aware navigation benchmarks to memory, prediction, and language-conditioned planning.
ECCV 2026 Corresponding author
Towards Unified World Models for Visual Navigation via Memory-Augmented Planning and Foresight
Combines memory, future prediction, and planning to navigate through unfamiliar environments.
IROS 2026 Corresponding author
HA-VLN 2.0: An Open Benchmark and Leaderboard for Human-Aware Navigation in Discrete and Continuous Environments with Dynamic Multi-Human Interactions
Provides a benchmark and evaluation platform for navigation amid dynamic interactions with multiple people.
NeurIPS 2024 · Datasets and Benchmarks Spotlight Co-first authorCorresponding author
Human-Aware Vision-and-Language Navigation
Introduces human-aware vision-and-language navigation, bringing moving people into embodied navigation tasks.
Findings of ACL 2026 Corresponding author
GoViG: Goal-Conditioned Visual Navigation Instruction Generation via Multimodal Reasoning
Generates goal-conditioned navigation instructions through multimodal reasoning about routes and visual observations.
Visual perception & retrieval
Contributions to crowd counting, human action and pose understanding, streaming perception, and cross-domain visual retrieval.
CVPR 2022 First author
Rethinking Spatial Invariance of Convolutional Networks for Object Counting
Revisits spatial invariance in convolutional networks to improve object counting across varied scenes.
ICCV 2019 OralCo-first author
Learning Spatial Awareness to Improve Crowd Counting
Introduces spatially aware density estimation and a pixel-level discrepancy loss to improve crowd counting in dense, noisy scenes.
CVPR 2024 Corresponding author
BlockGCN: Redefine Topology Awareness for Skeleton-Based Action Recognition
Preserves skeletal topology with geometric encodings and efficient graph convolutions for human action recognition.
IJCAI 2023 Co-first author
HDFormer: High-Order Directed Transformer for 3D Human Pose Estimation
Models higher-order relations among joints and bones for efficient, accurate 3D human pose estimation.
IJCAI 2023 Corresponding author
DAMO-StreamNet: Optimizing Streaming Perception in Autonomous Driving
Optimizes streaming perception for autonomous driving, accounting for both detection quality and inference latency.
CVPR 2017 First author
Video2Shop: Exact Matching Clothes in Videos to Online Shopping Images
Introduces cross-domain video-to-product retrieval, matching clothing in video sequences to the same items in online shops.
ACM Multimedia 2018 First author
Learning to Transfer: Generalizable Attribute Learning with Multitask Neural Model Search
Combines multitask learning and neural architecture search to transfer attribute representations across animals, objects, and scenes.
IEEE Transactions on Multimedia 2017 First author
Video eCommerce++: Toward Large Scale Online Video Advertising
Extends video advertising to large-scale deployment through incremental semantic modeling, cross-domain preference learning, and distributed optimization.
ACM Multimedia 2016 ACM SCF Best Student Paper AwardFirst author
Video eCommerce: Towards Online Video Advertising
Connects video semantics, shopping preferences, and viewing behavior to recommend products at relevant moments in a video.
Multimodal reasoning & generation
Models and datasets for understanding charts, situations, emotions, and events, and for generating images and human motion.
ICCV 2023 First author
ChartReader: A Unified Framework for Chart Derendering and Comprehension without Heuristic Rules
Unifies chart structure recovery and chart understanding without hand-crafted heuristic rules.
ACM Multimedia 2022 First author
GSRFormer: Grounded Situation Recognition Transformer with Alternate Semantic Attention Refinement
Jointly reasons about actions, participants, and their locations to recognize grounded visual situations.
NeurIPS 2024 Co-first authorCo-corresponding author
Emotion-LLaMA: Multimodal Emotion Recognition and Reasoning with Instruction Tuning
Combines audio, visual, and textual evidence for instruction-following emotion recognition and reasoning.
ICLR 2025 Corresponding author
MetaDesigner: Advancing Artistic Typography through AI-Driven, User-Centric, and Multilingual WordArt Synthesis
Integrates user intent, glyph design, and texture generation for multilingual artistic typography.
IEEE TPAMI 2025 · Early access Co-first author
Combo: Co-Speech Holistic 3D Human Motion Generation and Efficient Customizable Adaptation in Harmony
Generates coordinated facial expressions and body motion from speech, with efficient adaptation to new identities and emotions.
CVPR 2025
StableAnimator: High-Quality Identity-Preserving Human Image Animation
Generates human animation from a reference image while preserving the subject’s identity.
EMNLP 2024 · Industry Track Co-first author
SHIELD: LLM-Driven Schema Induction for Predictive Analytics in EV Battery Supply Chain Disruptions
Uses language models to induce event schemas for predicting disruptions in electric-vehicle battery supply chains.
NAACL 2025
ProMQA: Question Answering Dataset for Multimodal Procedural Activity Understanding
Introduces multimodal question answering for understanding procedural activities and their execution.
ACL 2026 Corresponding author
Sign-Language Datasets at Scale: A Comprehensive Survey on Resources, Benchmarks, and Annotation Standards
Organizes 120 datasets across 35 sign languages and introduces a 24-field datasheet for consistent documentation and evaluation.
CVPR 2025 · Anti-UAV Workshop Best Paper Award Corresponding author
Securing the Skies: A Comprehensive Survey on Anti-UAV Methods, Benchmarking, and Future Directions
Systematizes anti-UAV methods, sensing datasets, evaluation practices, and open research challenges.
Efficient & robust learning
Faster inference, stronger representations, and reliable adaptation of foundation models.
ICLR 2026 OralCo-corresponding author
Overcoming Joint Intractability with Lossless Hierarchical Speculative Decoding
Accelerates language-model inference through hierarchical speculative decoding while preserving the target output distribution.
NeurIPS 2025 OralCorresponding author
MaxSup: Overcoming Representation Collapse in Label Smoothing
Addresses representation collapse caused by label smoothing while retaining the benefits of regularization.
NeurIPS 2024 Co-corresponding author
Towards Calibrated Robust Fine-Tuning of Vision-Language Models
Develops fine-tuning methods that improve both prediction accuracy and confidence calibration under distribution shift.
IEEE TPAMI 2026
Robust Adaptation of Foundation Models with Black-Box Visual Prompting
Adapts black-box foundation models through visual prompts, with parameter-efficient learning and robustness analysis.