ChartCF achieves strong chart understanding performance in VLMs using significantly less training data by generating code-based counterfactuals, selecting similar samples, and performing multimodal preference optimization.
Multiple data values are modified (e.g., Sun B Solar Flares from 6.8 to 9.7), resulting in a different final sum while maintaining similar chart appearance
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Learning More from Less: Exploiting Counterfactuals for Data-Efficient Chart Understanding
ChartCF achieves strong chart understanding performance in VLMs using significantly less training data by generating code-based counterfactuals, selecting similar samples, and performing multimodal preference optimization.