HRScene: A High-Resolution Image Understanding Benchmark

HRScene is a unified benchmark for high-resolution image understanding, incorporating 25 real-world and 2 synthetic datasets. The collection includes images with resolutions ranging from 1,024 × 1,024 to 35,503 × 26,627, covering 25 scenarios. It was collected and re-annotated by 10 graduate-level annotators and was last updated on April 28, 2025.

Use Cases

• Benchmarking model performance on high-resolution images based on the wide range of resolutions provided.
• Training scene understanding models based on the 25 covered real-world scenarios.
• Developing diagnostic tools for medical imaging based on the inclusion of radiology and microscopic datasets.
• Evaluating model robustness across diverse image scales based on the synthetic diagnostic datasets.

Strengths

• Includes 25 real-world datasets and 2 synthetic diagnostic datasets.
• Covers a wide resolution range from 1,024 × 1,024 to 35,503 × 26,627.
• Annotations were performed by 10 graduate-level annotators.
• Encompasses 25 distinct scenarios, including medical domains.

Limitations

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

Provenance

Source

Wenliang04 on Hugging Face.

Collection Method

Collected and re-annotated from 25 existing datasets.

Freshness

Last updated 2025-04-28 02:17:45; freshness should be verified.