ImagineAgent uses generative world modeling and tool-augmented reinforcement learning to reach state-of-the-art open-vocabulary HOI performance on SWIG-HOI and HICO-DET while using only 36.7% of the training data required by prior methods.
However, existing methods struggle with long-sequence hallucination (Chen et al., 2025b) and limited cross-modal interaction
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
-
What if Agents Could Imagine? Reinforcing Open-Vocabulary HOI Comprehension through Generation
ImagineAgent uses generative world modeling and tool-augmented reinforcement learning to reach state-of-the-art open-vocabulary HOI performance on SWIG-HOI and HICO-DET while using only 36.7% of the training data required by prior methods.