Selected work on spoof detection, forgery analysis, and explainable multimodal reasoning.
Reasoning
FAS-R1 is a reasoning-oriented multimodal large language model for unified face anti-spoofing, covering authenticity classification, attack-type recognition, and spoof-region localization.
Unified Detection
UniShield introduces a knowledge-grounded multimodal reasoning framework for unified face attack detection across both physical spoofing and digital forgery scenarios.
Explainability
FaceShield proposes an explainable multimodal large language model framework for face anti-spoofing, covering binary classification, attack type recognition, reasoning, and region localization.
Benchmark
SHIELD is an evaluation benchmark for multimodal large language models on both face anti-spoofing and face forgery detection tasks.
Domain Adaptation
This work addresses the cross-domain generalization problem in face anti-spoofing by introducing secondary disentanglement and progressive liveness feature alignment.
Selected work on AMS datasets, benchmarking, knowledge-enhanced design, and component generation.
Dataset
AMSNet builds a large-scale paired dataset of AMS schematics and netlists to support multimodal understanding and automatic design for analog and mixed-signal circuits.
Dataset
AMSnet 2.0 expands the AMS database with AI-based segmentation for net detection and richer positional and digital representations of circuit schematics.
Knowledge Graph
AMSnet-KG extends AMS netlist data with a circuit knowledge graph and retrieval-augmented generation to support LLM-based reasoning and auto-design for AMS circuits.
Benchmark
AMSbench is a comprehensive benchmark for evaluating multimodal large language models on AMS circuit perception, analysis, and design tasks.
Generation
This work presents an agentic recognition-and-generation pipeline for electronic component symbols and footprints, together with a scalable database for automated PCB design.