CORTEXA
← Browse

Ye Huang

2 papers indexed

arxivcs.CV2026-07-17

FVAttn: Adaptive Sparse Attention with Runtime Load Balancing for Video Generation

Hao Liu, Chenghuan Huang, Ye Huang, Zhiying Wen, Hao Liu, Mohan Zhang, et al.

Video Diffusion Transformers process long spatio-temporal sequences, making self-attention the main bottleneck in high-resolution video generation. Training-free sparse attention reduces this cost, but adaptive Top-$p$ routing creates uneven per-head workloads under multi-GPU seq…

View free PDFSource page
arxivcs.CVcs.AIcs.RO2026-06-26

PhysisForcing: Physics Reinforced World Simulator for Robotic Manipulation

Peiwen Zhang, Yufan Deng, Shangkun Sun, Juncheng Ma, Duomin Wang, Jonas Du, et al.

Video generation models have emerged as a promising paradigm for embodied world simulation. However, both general-domain video generators and robot-specific data fine-tuned models can still produce physically implausible manipulations, including discontinuous motion trajectories…

View free PDFSource page