Search papers, labs, and topics across Lattice.
Affiliation:
4
0
6
Temporal degradation in video diffusion isn't just caused by weak cross-frame attention鈥攊t is driven by an underlying spectral imbalance between local diffuseness and global diversity that can be selectively repaired at inference time without retraining.
SkillTrace redefines skill composition for LLM agents, achieving unprecedented success rates by leveraging a structured query-skill graph.
Video LLMs can get a free performance boost by using ST-GridPool, a novel technique that enhances visual token representations without any additional training.
Instead of just pruning redundant tokens, ST-SimDiff dramatically cuts MLLM video processing costs by intelligently preserving tokens representing *changes* in the video.
