MATH-Vision is a new benchmark of 3,040 visual mathematical competition problems that reveals substantial gaps between large multimodal models and human performance in mathematical reasoning.
Each cloud seems to represent a puzzle or problem where the numbers are related to each other in some way, possibly through mathematical operations
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2024 1verdicts
ACCEPT 1representative citing papers
citing papers explorer
-
Measuring Multimodal Mathematical Reasoning with MATH-Vision Dataset
MATH-Vision is a new benchmark of 3,040 visual mathematical competition problems that reveals substantial gaps between large multimodal models and human performance in mathematical reasoning.