Selected work

Representative Papers

SPARK-IL: Spectral Retrieval-Augmented RAG for Knowledge-driven Deepfake Detection via Incremental Learning

Apr 04, 2026

This work addresses the limited generalization of existing AI-generated image detection methods to unseen generative models by proposing a novel incremental learning framework that integrates dual-path spectral analysis with retrieval-augmented generation (RAG). The approach employs four-band Fourier decomposition to extract frequency-domain features, combines a partially frozen ViT-L/14 encoder with Kolmogorov–Arnold Network (KAN)-based mixture-of-experts to model band-specific characteristics, and incorporates elastic weight consolidation for continual learning. Notably, it introduces, for the first time, a synergy between spectral consistency priors and RAG, leveraging a Milvus vector database for knowledge retrieval to enhance discriminative robustness. Evaluated on the UniversalFakeDetect benchmark encompassing 19 generative models, the method achieves an average accuracy of 94.6%, substantially outperforming current state-of-the-art techniques.

0 citationsRead paper

Smartphone Exergames with Real-Time Markerless Motion Capture: Challenges and Trade-offs

Jul 09, 2025

Unmarked motion capture for mobile health and rehabilitation faces inherent trade-offs between accuracy and latency, alongside insufficient adaptability to user interaction. Method: This study proposes a lightweight, real-time exergame framework tailored for smartphones, integrating mobile-optimized AI pose estimation, low-latency video streaming, and adaptive gamified interaction design to achieve sub-second motion tracking on standard Android/iOS devices. Contributions/Results: (1) First systematic validation of a camera-only smartphone solution for home-based rehabilitation, demonstrating both technical feasibility and high user acceptability; (2) A co-optimization strategy for accuracy and real-time performance, reducing inference latency by 32% over baseline models while maintaining >92% keypoint detection accuracy; (3) A scalable, hardware-free remote rehabilitation prototype system that significantly improves user engagement and accessibility.

0 citationsRead paper

Recent Advances in Medical Imaging Segmentation: A Survey

May 14, 2025

Medical image segmentation faces fundamental challenges including data scarcity, high annotation costs, poor cross-modal and cross-domain generalizability, and stringent privacy constraints. To address these, this work systematically reviews over 200 state-of-the-art publications and—uniquely—integrates perspectives from generative AI (e.g., diffusion models) and foundation models into a clinically oriented evaluation framework. We delineate adaptation pathways for foundation models in medical segmentation, identify critical bottlenecks (e.g., domain misalignment, computational overhead), and propose lightweight fine-tuning strategies—including visual prompting, multimodal fusion, and self-supervised pretraining. Our contributions include a continuously updated, open-source knowledge repository on GitHub, featuring a structured technical roadmap that bridges algorithmic innovation with clinical translation. This resource supports both methodological development and real-world deployment of segmentation models in healthcare settings.

0 citationsRead paper

Recent publications

Latest Papers

SPARK-IL: Spectral Retrieval-Augmented RAG for Knowledge-driven Deepfake Detection via Incremental Learning

Apr 04, 2026

This work addresses the limited generalization of existing AI-generated image detection methods to unseen generative models by proposing a novel incremental learning framework that integrates dual-path spectral analysis with retrieval-augmented generation (RAG). The approach employs four-band Fourier decomposition to extract frequency-domain features, combines a partially frozen ViT-L/14 encoder with Kolmogorov–Arnold Network (KAN)-based mixture-of-experts to model band-specific characteristics, and incorporates elastic weight consolidation for continual learning. Notably, it introduces, for the first time, a synergy between spectral consistency priors and RAG, leveraging a Milvus vector database for knowledge retrieval to enhance discriminative robustness. Evaluated on the UniversalFakeDetect benchmark encompassing 19 generative models, the method achieves an average accuracy of 94.6%, substantially outperforming current state-of-the-art techniques.

0 citationsRead paper

Smartphone Exergames with Real-Time Markerless Motion Capture: Challenges and Trade-offs

Jul 09, 2025

Unmarked motion capture for mobile health and rehabilitation faces inherent trade-offs between accuracy and latency, alongside insufficient adaptability to user interaction. Method: This study proposes a lightweight, real-time exergame framework tailored for smartphones, integrating mobile-optimized AI pose estimation, low-latency video streaming, and adaptive gamified interaction design to achieve sub-second motion tracking on standard Android/iOS devices. Contributions/Results: (1) First systematic validation of a camera-only smartphone solution for home-based rehabilitation, demonstrating both technical feasibility and high user acceptability; (2) A co-optimization strategy for accuracy and real-time performance, reducing inference latency by 32% over baseline models while maintaining >92% keypoint detection accuracy; (3) A scalable, hardware-free remote rehabilitation prototype system that significantly improves user engagement and accessibility.

0 citationsRead paper

Recent Advances in Medical Imaging Segmentation: A Survey

May 14, 2025

Medical image segmentation faces fundamental challenges including data scarcity, high annotation costs, poor cross-modal and cross-domain generalizability, and stringent privacy constraints. To address these, this work systematically reviews over 200 state-of-the-art publications and—uniquely—integrates perspectives from generative AI (e.g., diffusion models) and foundation models into a clinically oriented evaluation framework. We delineate adaptation pathways for foundation models in medical segmentation, identify critical bottlenecks (e.g., domain misalignment, computational overhead), and propose lightweight fine-tuning strategies—including visual prompting, multimodal fusion, and self-supervised pretraining. Our contributions include a continuously updated, open-source knowledge repository on GitHub, featuring a structured technical roadmap that bridges algorithmic innovation with clinical translation. This resource supports both methodological development and real-world deployment of segmentation models in healthcare settings.

0 citationsRead paper