CVPR

UPLiFT: Efficient Pixel-Dense Feature Upsampling with Local Attenders

UPLiFT：高效像素密集特征上采样与局部关注者

Matthew Walmer, Saksham Suri, Anirud Aggarwal, Abhinav Shrivastava

机构 * University of Maryland, College Park（马里兰大学学院公园分校） ； Meta 纠错

AI总结 UPLiFT通过高效局部关注者操作符实现低成本的像素密集特征上采样，优于现有方法。

Comments Accepted to CVPR 2026 as conference paper. Website: see https://www.cs.umd.edu/~mwalmer/uplift/. Code and Models: see https://github.com/mwalmer-umd/UPLiFT

OmniMotion-X: Versatile Multimodal Whole-Body Motion Generation

OmniMotion-X：通用多模态全身运动生成

Guowei Xu, Yuxuan Bian, Ailing Zeng, Zhuo Chen, Mingyi Shi, Shaoli Huang, Wen Li, Lixin Duan, Qiang Xu

AI总结 介绍用于全身人类运动生成的OmniMotion-X框架，利用自回归扩散变换器，提出参考运动等方法，构建数据集并采用新训练策略，超越现有方法，在多模态任务中表现出色，可交互式生成逼真连贯可控的长时间运动。

Comments Accepted to CVPR 2026 Findings