A binary-search style reward that halves candidate objects each round improves goal-oriented visual dialogue accuracy and reduces question repetition.
In Findings of the Association for Computational Linguistics: EMNLP 2021
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Divide-and-Conquer: Tree-structured Strategy with Answer Distribution Estimator for Goal-Oriented Visual Dialogue
A binary-search style reward that halves candidate objects each round improves goal-oriented visual dialogue accuracy and reduces question repetition.