Explore · 每周精选
发现最受关注的研究论文,追踪研究趋势,订阅感兴趣的期刊与关键词。
36 篇论文 · 按点赞排序
Hao Zhang, Yao Feng, Peter Kulits +3 authors
A system generating realistic 3D avatars from text descriptions using a compositional model with traditional 3D meshes for the face and neural radiance fields (NeRF) for hair, accessories, and clothing.
Chen Geng, Hong-Xing Yu, Sharon Zhang +2 authors
A hybrid approach combining an auto-regressive model and optimization algorithm is proposed for inferring a shade tree representation from images, enabling intuitive shading editing.
Xinyi Wang, John Wieting, Jonathan H. Clark
FIAT, a new learning paradigm, combines in-context learning and fine-tuning to improve performance across multilingual tasks using different model sizes.
Yao Feng, Weiyang Liu, Timo Bolkart +3 authors
Disentangled Avatars (DELTA) uses hybrid explicit-implicit 3D representations to model disentangled human body, clothing, face, and hair from monocular videos, enabling applications like virtual try-on and hairstyle transfer.
Lingyu Gao, Aditi Chaudhary, Krishna Srinivasan +3 authors
In-context learning benefits from selecting demonstrations that resolve label ambiguity and include examples on the decision boundary, leveraging the LLM's existing knowledge.
Rishav Hada, Varun Gumma, Adrian de Wynter +5 authors
LLM-based evaluators show promise for scaling multilingual NLP tasks evaluation but may have biases toward higher scores and require calibration with native speaker judgments.
北京市昌平区探索星信息技术及软件开发工作室
京ICP备2026059466号