Publications

Tags

DiffRGD: A Training-Free Diffusion Guidance Through Riemannian Gradient Descent

Jia-Wei Liao, Li-Xuan Peng, Mei-Heng Yueh, Min Sun, Cheng-Fu Chou, Jun-Cheng Chen

In submission

M-ErasureBench: A Comprehensive Multimodal Evaluation Benchmark for Concept Erasure in Diffusion Models

Ju-Hsuan Weng*, Jia-Wei Liao*, Cheng-Fu Chou, Jun-Cheng Chen

WACV 2026

TL;DR

M-ErasureBench is a multimodal benchmark revealing that existing diffusion concept-erasure methods fail beyond text prompts and introduces an inference-time module that significantly improves erasure robustness without retraining.

Zero-shot Geometry-Aware Diffusion Guidance for Music Restoration

Jia-Wei Liao, Pin-Chi Pan, Li-Xuan Peng, Sheng-Ping Yang, Yen-Tung Yeh, Cheng-Fu Chou, Yi-Hsuan Yang

NeurIPS 2025 AI4Music Workshop

TL;DR

Diffusion Geodesic Guidance (DGG) is a zero-shot geometry-aware diffusion guidance method that updates samples along hyperspherical geodesics to preserve the model prior while improving music restoration quality without retraining.

BEVAN: Bilateral Efficient Visual Attention Network for Real-Time Semantic Segmentation

Ping-Mao Huang, I-Tien Chao, Ping-Chia Huang, Jia-Wei Liao, Yung-Yu Chuang

ICIP 2025

TL;DR

BEVANet is a bilateral large-kernel attention network that achieves state-of-the-art real-time semantic segmentation by adaptively expanding receptive fields and fusing semantic, structural, and boundary features efficiently.

DiffQRCoder: Diffusion-based Aesthetic QR Code Generation with Scanning Robustness Guided Iterative Refinement

Jia-Wei Liao, Winston Wang*, Tzu-Sian Wang*, Li-Xuan Peng*, Ju-Hsuan Weng, Cheng-Fu Chou, Jun-Cheng Chen

WACV 2025

TL;DR

DiffQRCoder is a training-free diffusion framework that generates visually appealing QR codes while preserving high scanning robustness through geometry-aware perceptual guidance and iterative refinement.

Pixel Is Not A Barrier: An Effective Evasion Attack for Pixel-Domain Diffusion Models

Chun-Yen Shih*, Li-Xuan Peng*, Jia-Wei Liao, Ernie Chu, Cheng-Fu Chou, Jun-Cheng Chen

AAAI 2025

TL;DR

AtkPDM is a feature-space adversarial attack framework that crafts imperceptible perturbations to protect images from unauthorized diffusion-based editing by disrupting UNet representations while preserving visual fidelity.

Distribution Discrepancy and Feature Heterogeneity for Active 3D Object Detection

Huang-Yu Chen, Jia-Fong Yeh, Jia-Wei Liao, Pin-Hsuan Peng, Winston H Hsu

CoRL 2024

TL;DR

DDFH is an active learning framework for LiDAR-based 3D object detection that selects the most informative samples by jointly modeling distribution discrepancy and feature heterogeneity to reduce annotation cost while improving detection performance.

An UNet-Based Brain Tumor Segmentation Framework via Optimal Mass Transportation Pre-processing

Jia-Wei Liao, Tsung-Ming Huang, Tiexiang Li, Wen-Wei Lin, Han Wang, Shing-Tung Yau

MICCAI 2022 Brainlesion Workshop

TL;DR

This paper proposes a two-phase UNet-based brain tumor segmentation framework that uses optimal mass transportation to enlarge tumor regions and enhance data diversity, significantly improving MRI segmentation accuracy and robustness.