HY-Himmel improves Video-MME accuracy from 61.2% to 63.5% with 3.6x fewer tokens by separating semantic grounding on sparse anchors from compressed-domain motion encoding on dense intervals.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
HY-Himmel Technical Report: Hierarchical Interleaved Multi-stream Motion Encoding for Long Video Understanding
HY-Himmel improves Video-MME accuracy from 61.2% to 63.5% with 3.6x fewer tokens by separating semantic grounding on sparse anchors from compressed-domain motion encoding on dense intervals.