Yihang Lou

Global consistency in Text-to-SVG generation can be achieved without additional training, thanks to a novel adaptive sampling strategy that intelligently navigates stroke decisions.

TUE-Detector outperforms traditional methods by using a multi-layered language model to invoke specialized tools for uncovering subtle artifacts in AI-generated videos.

Ablating just 5% of specialized CoRe heads leads to a dramatic drop in multimodal reasoning performance, revealing their critical role in MLLMs.

ToolFG revolutionizes fine-grained image classification by enabling MLLMs to autonomously leverage external tools for enhanced reliability and accuracy.