---
language:
  - en
tags:
  - remote-sensing
  - image-quality-assessment
  - benchmark
  - visual-question-answering
task_categories:
  - visual-question-answering
  - image-classification
pretty_name: SenseBench
license: cc-by-4.0
dataset_info:
  features:
  - name: id
    dtype: string
  - name: image_1
    dtype: image
  - name: image_2
    dtype: image
  - name: question
    dtype: string
  - name: answer
    dtype: string
  splits:
  - name: train
    num_examples: 10815
  configs:
  - config_name: default
    data_files:
    - split: train
      path: questions.jsonl
---

# SenseBench

A benchmark for remote sensing low-level visual perception and description in large vision-language models.

- **10,815** samples, **14,330** images
- 4 question types: whether / what / how / description

## Task Types

| Category | Task | Samples | Description |
|----------|------|---------|-------------|
| Perception | whether | 3,800 | Yes/No binary questions |
| Perception | what | 3,800 | Single-choice (A/B/C/D) |
| Perception | how | 2,015 | Severity rating (A/B/C) |
| Description | description | 1,200 | Open-ended quality description |

## Directory Structure

```
├── Perception/
│   ├── single/
│   └── paired/
├── Description/
│   ├── single/
│   └── paired/
└── questions.jsonl
```