ReFine3D uses selective layer tuning, multi-view consistency regularization, LLM-generated text diversity, point-rendered supervision, and confidence-weighted test-time augmentation to improve domain generalization in 3D LMMs by 1-3% on benchmarks.
arXiv preprint arXiv:2603.23730 , year=
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
-
Domain Generalizable Adaptation of 3D Vision-Language Models via Regularized Fine-Tuning
ReFine3D uses selective layer tuning, multi-view consistency regularization, LLM-generated text diversity, point-rendered supervision, and confidence-weighted test-time augmentation to improve domain generalization in 3D LMMs by 1-3% on benchmarks.