Dataset · DreamMr
HR-Bench benchmark for high-resolution photo analysis
DreamMr/HR-Bench
English question-answer set to test the ability of multimodal models to recognize details in images above 2K.
What's inside
The benchmark is designed to evaluate multimodal language models (MLLMs) on ultra-high-resolution images (above 2K). It contains 1 to 10 thousand English examples divided into two subtasks: single-object fine-grained perception (FSP) and multi-object cross-object relationship perception (FCP).
What you can do with it
- Testing multimodal LLMs (VQA) for noticing fine details in large images
- Comparing high-resolution image processing methods without additional training
- Tasks
- Visual question answering
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 31.5K
- Likes
- 19
How to load it
from datasets import load_dataset
ds = load_dataset("DreamMr/HR-Bench")