Search papers, labs, and topics across Lattice.
1Zhejiang University
7
0
9
Achieving over 90% performance retention with a staggering 20x KV cache compression could redefine efficiency in long-context audio inference.
A semantic re-keying strategy in speculative decoding can boost accepted draft lengths by up to 29% while achieving 4.4x faster decoding speeds.
Bridging the reasoning gap, X$^3$-OPD enables audio-language models to outperform their text-based counterparts in logical reasoning tasks.
HierCAD achieves unprecedented fidelity in CAD generation by aligning structural reasoning with geometric parameters, setting a new benchmark for text-to-CAD systems.
Achieving a harmonious balance between temporal consistency and editability in video editing could redefine standards in text-guided video manipulation.
Automatically constructed data can dramatically enhance the temporal localization abilities of audio models, overcoming the limitations of manual annotation.
ROGLE achieves a breakthrough in Text-Based Person Search by automatically generating fine-grained supervision, outperforming existing models on challenging long-form queries.
