osv5m
VerifiedLarge-scale street view dataset for global visual geolocation research.
What is osv5m?
OpenStreetView-5M consists of street view imagery collected for training and evaluating models on worldwide visual geolocation.
It supports machine learning research in computer vision, particularly tasks that require inferring geographic position from image content.
What you can build with osv5m
Train Visual Geolocation Models
Develop deep learning models that predict latitude and longitude from street-level photos using the 5M image set.
Benchmark Image-Based Localization
Test and compare retrieval or regression algorithms on a globally distributed street-view collection.
Create Location Inference Tools
Fine-tune models for apps that estimate geographic position from user photos at worldwide scale.
Load osv5m
from datasets import load_dataset
ds = load_dataset("osv5m/osv5m")- 1Install the Hugging Face datasets library with pip
- 2Import load_dataset from the datasets package
- 3Load the dataset with load_dataset('osv5m/osv5m')
- 4Select the train split containing images and coordinates
- 5Preprocess images and labels for your geolocation pipeline
osv5m: pros & cons
Pros
- +5 million street-level images
- +Global geographic coverage
- +Purpose-built for visual geolocation
- +Directly loadable via Hugging Face
Cons
- –Large storage and compute requirements
- –No additional annotations beyond location
- –Regional image distribution may be uneven
Frequently asked questions
A dataset of 5 million street-level images created for visual geolocation tasks by researchers at Imagine, LIGM, Ecole des Ponts.
User reviews
Verified reviews from the community shape this listing's rating.
Loading reviews…
Similar datasets
Other ai & machine learning options worth comparing.