MTabVQA is a new visual multi-table question answering benchmark, and fine-tuning VLMs on its instruction set improves their accuracy on it.
We utilized the EasyR1 framework7 for these experiments, training for a total of 270 steps
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
MTabVQA: Evaluating Multi-Tabular Reasoning of Language Models in Visual Space
MTabVQA is a new visual multi-table question answering benchmark, and fine-tuning VLMs on its instruction set improves their accuracy on it.