Medical VLM Robustness Benchmark for Image Degradation

18 clinically-relevant image degradation types are applied across 7 medical imaging modalities, each at mild and severe levels. The dataset benchmarks the robustness of medical Vision-Language Models (VLMs) against these degradations. It was created by jiyaoliufd and last updated in March 2026.

Use Cases

• Evaluating VLM performance on degraded medical images across 7 modalities like X-ray and MRI.
• Analyzing the impact of 18 specific degradation types, such as motion blur or noise, on visual question answering accuracy.
• Benchmarking model robustness by comparing results on mild versus severe levels of applied degradation.
• Training or fine-tuning VLMs to be more resilient to specific, clinically-observed image artifacts.

Strengths

• Covers 18 distinct, clinically-relevant image degradation types.
• Includes 7 different medical imaging modalities for broad applicability.
• Each degradation type is applied at two distinct severity levels (mild and severe).

Limitations

• The total number of samples is unknown, making scale assessment impossible.
• Specific column names and data structure are not provided in the input.
• Geographic and temporal coverage of the source medical images is unknown.

Provenance

Source

huggingface, author jiyaoliufd

Collection Method

Constructed as a benchmark by applying controlled degradations to medical images, likely for VLM evaluation.

Freshness

Last updated March 2026.