STER-VLM decomposes traffic captions into spatial and temporal parts, selects a few informative frames, and adds 72B-model reference hints, yielding a small combined validation gain and a 55.655 AI City Challenge Track 2 score.
Lora: Low-rank adaptation of large language models
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
STER-VLM: Spatio-Temporal With Enhanced Reference Vision-Language Models
STER-VLM decomposes traffic captions into spatial and temporal parts, selects a few informative frames, and adds 72B-model reference hints, yielding a small combined validation gain and a 55.655 AI City Challenge Track 2 score.