shanmuga12nivetha/university-1652

---
license: fair-noncommercial-research-license
language:
- en
tags:
- computer-vision
- geo-localization
- drone
- satellite-imagery
dataset_info:
features:
- name: image
dtype: string
- name: building_id
dtype: string
- name: view_type
dtype: string
- name: split_type
dtype: string
splits:
- name: train
num_bytes: 3176316
num_examples: 49644
- name: test
num_bytes: 6458377
num_examples: 96372
download_size: 1363677
dataset_size: 9634693
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
---

# University-1652: Drone-based Geo-localization Benchmark 🚁

<image-card alt="Sample" src="https://raw.githubusercontent.com/layumi/University1652-Baseline/master/docs/index_files/Data.jpg" ></image-card>

University-1652 is a multi-view dataset for drone-based geo-localization, annotating **1652 buildings across 72 universities** (ACM Multimedia 2020, [paper](https://arxiv.org/abs/2002.12186)). Cited in **50+ papers**, it supports **Drone → Satellite localization** and **Satellite → Drone navigation**.

## Dataset Structure
- **Splits**:
- Train: 50,218 images (drone, satellite, street, google; 33 universities)
- Test:
- query_drone: 37,855 images
- gallery_drone: 51,355 images
- query_street: 2,579 images
- gallery_street: 2,921 images
- query_satellite: 701 images
- gallery_satellite: 951 images
- 4K_drone: 12 images
- **Features**:
- `image`: Drone/satellite/street/4K_drone images
- `building_id`: Building identifier
- `view_type`: drone/satellite/street/drone_4k
- `split_type`: train/query/gallery
- **Size**: ~9.2GB (unzipped)

## Usage
```python
from datasets import load_dataset
ds = load_dataset("layumi/university-1652", split="train")
ds[0] # {'image': ..., 'building_id': '0001', 'view_type': 'drone', 'split_type': 'train'}