Fundus Imaging Benchmark For Multimodal Language Model Evaluation

620 data points comprise the Fundus-MMBench, a benchmark for evaluating ophthalmic multimodal large language models. It was created by MeteorElf and last updated in August 2025. The dataset supports research into clinical cognitive chain reasoning for positioning and diagnosis collaboration.

Use Cases

• Evaluate model performance on fundus image analysis tasks using the benchmark's structured test cases.
• Train models for clinical cognitive chain reasoning by leveraging the dataset's positioning-diagnosis collaboration framework.
• Assess multimodal language model capabilities in interpreting ophthalmic data and generating diagnostic insights.

Strengths

• Contains 620 specifically curated data points for benchmark evaluation.
• Designed for a focused research purpose: evaluating ophthalmic MLLMs through clinical reasoning chains.

Limitations

• Small scale with only 620 total data points, limiting statistical power for some analyses.
• Limited to academic research use only, restricting commercial or clinical deployment applications.
• Specific column structure and data features are not publicly detailed in the provided description.

Provenance

Source

huggingface, author MeteorElf

Collection Method

Constructed as a benchmark for a specific research paper on ophthalmic MLLMs.

Freshness

Last updated on 2025-08-08.