CosPlace: Large-Scale Visual Geo-localization for Place Recognition

Developed for the CVPR 2022 paper 'Rethinking Visual Geo-localization for Large-Scale Applications', CosPlace offers a framework for training deep learning descriptors. It facilitates large-scale image retrieval by treating place recognition as a classification task across geographic locations.

Use Cases

• Training deep learning descriptors for visual place recognition
• Benchmarking image retrieval algorithms using geolocalization labels
• Developing Pytorch-based models for large-scale place recognition

Strengths

• Peer-reviewed methodology from CVPR 2022
• MIT licensed for open research
• Includes Pytorch-compatible descriptors for place recognition

Limitations

• Specific row counts and file sizes are not provided in the repository metadata
• Requires external download or script execution to reconstruct the full dataset

Provenance

Source

CVPR 2022 paper 'Rethinking Visual Geo-localization for Large-Scale Applications'

Collection Method

Likely aggregated from existing street-view or mapping imagery sources as per the paper's methodology

Freshness

Last updated February 2026; based on 2022 research.

Geography

Global/Large-scale urban environments