VEME, a dual-memory cross-modal alignment framework built on Qwen-2.5-VL, reports modest gains on VLN-CE and VSI-Bench that are contradicted by its own internal numbers.
X.; Savva, M.; Halber, M.; Funkhouser, T.; and Nie ner, M
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
-
Beyond Pixels: Introducing Geometric-Semantic World Priors for Video-based Embodied Models via Spatio-temporal Alignment
VEME, a dual-memory cross-modal alignment framework built on Qwen-2.5-VL, reports modest gains on VLN-CE and VSI-Bench that are contradicted by its own internal numbers.