المفاهيم الأساسية
TriAdapter Multi-Modal Learning (TAMM) enhances 3D shape understanding by effectively leveraging image and text modalities.
الإحصائيات
TAMM improves zero-shot classification accuracy from 46.8% to 50.7% on Objaverse-LVIS.
TAMM enhances 5-way 10-shot linear probing classification accuracy from 96.1% to 99.0% on ModelNet40.
اقتباسات
"TAMM significantly enhances 3D shape understanding by better exploiting the image modality."
"Our proposed TAMM consistently enhances 3D representations for a wide range of 3D encoder architectures."