Training VLMs to point via text induces serial processing that eliminates binding errors and enables compositional generalization on multi-object tasks.
Evaluating compositional scene understanding in multimodal generative models.arXiv preprint arXiv:2503.23125,
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
-
Binding Visual Features Point by Point
Training VLMs to point via text induces serial processing that eliminates binding errors and enables compositional generalization on multi-object tasks.