Hong-Han Shuai

Density-Assisted Adaptive Alignment Network for Video Individual Counting

From Global Alignment to Local Grounding: Zero-Shot Chinese Character Recognition with Radical Verification

This work proposes a global-to-local two-stage framework for zero-shot Chinese character recognition that achieves state-of-the-art performance under the character-level zero-shot setting, and shows that the explicit geometric priors and radical-level verification provide complementary improvements.

Look Before You Judge: Training-Free Region Mining for Grounded and Explainable Deepfake Detection

Look Before You Judge, a training-free framework that formulates explainable deepfake detection as a sequential evidence acquisition process, and improves detection accuracy by up to 12.8%, reduces CHAIR by up to 33.4% and hallucination rate by up to 21.3%, and outperforms representative training-free decoding and atte...

HalluPeer: A Taxonomy-driven Benchmark for Detecting Hallucinations in Scientific Peer Reviews

HalluPeer is introduced, a benchmark for detecting hallucinations in scientific peer reviews, providing aligned triples of paper content, human-written reviews, and hallucination-injected reviews, annotated for detection, classification, and localization.

Uncertainty-Guided Adaptive Knowledge Distillation for Lightweight Cross-Domain Object Detection

This article introduces Uncertainty-Guided Adaptive Knowledge Distillation (UGAKD), a novel framework designed to enhance UDA while simultaneously reducing model size through targeted knowledge distillation and proposing a two-stage, difficulty-aware training scheme to facilitate learning.

RADIANCE: Relative Adaptive Denoising with IP-Adapter for Novel Concept Enhancement

RADIANCE is proposed, a training-free framework that treats inference as a closed-loop feedback process that consistently enhances compositional alignment and perceptual quality over state-of-the-art baselines.