OddGridBench: Visual Discrepancy Sensitivity Benchmark for Multimodal LLMs

OddGridBench is a benchmark dataset designed to evaluate the fine-grained visual discrepancy sensitivity of Multimodal Large Language Models. The dataset and associated paper were accepted for presentation at CVPR 2026. It was created by the author 'wwwtttjjj' and is hosted on HuggingFace.

Use Cases

• Benchmarking model performance on fine-grained visual difference detection based on the described task.
• Analyzing failure modes of Multimodal LLMs in visual reasoning tasks.
• Training or fine-tuning models to improve sensitivity to subtle visual discrepancies.

Strengths

• Dataset is associated with a peer-reviewed paper accepted to CVPR 2026, suggesting academic rigor.
• The repository contains official evaluation code, which may support reproducibility.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

HuggingFace user 'wwwtttjjj'.

Collection Method

Created for the research paper 'OddGridBench: Exposing the Lack of Fine-Grained Visual Discrepancy Sensitivity in Multimodal Large Language Models'.

Freshness

Last updated 2026-06-02 09:27:38; freshness should be verified.