Adding ground-truth object localization, tracking, and relation annotations as context improves VideoLLM answers, but the models never generate these steps themselves in the experiments.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
CoTasks: Chain-of-Thought based Video Instruction Tuning Tasks
Adding ground-truth object localization, tracking, and relation annotations as context improves VideoLLM answers, but the models never generate these steps themselves in the experiments.