Sign in to view source links and access this dataset
Description
870 expert-validated editing triplets form the core of ChartSync, a benchmark for evaluating Visuo-Logical Cascading Editing (VLCE) in statistical chart images. The dataset, created by JiakangYu, includes 9 chart categories and 4 task types, with 235 geometry-coupled VLCE instances. It was last updated on July 11, 2026.
Use Cases
Train models for visuo-logical cascading editing based on the 870 expert-validated triplets.
Benchmark multimodal AI performance on chart understanding across 9 chart categories.
Evaluate model capabilities on 4 distinct chart editing task types.
Study geometry-coupled reasoning using the 235 specific VLCE instances.
Strengths
870 expert-validated editing triplets provide a substantial foundation for model training and evaluation.
Covers 9 distinct chart categories and 4 task types, suggesting diversity in editing scenarios.
Includes 235 geometry-coupled VLCE instances, a specific and likely challenging subset for reasoning.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Row count is unknown, which may limit suitability assessment for large-scale training.
Dataset size and file formats are unspecified, complicating download and storage planning.
Provenance
Source
huggingface
Collection Method
Likely created as a research benchmark; details on data gathering are not provided.
Freshness
Last updated 2026-07-11 13:28:14; freshness should be verified.
License is unknown; users should verify terms before use.