GoPro 360 Panorama Dataset - Index & Documentation

Quick Links: extracted | annotations | annotation galleries | gis | buildings | panos | triage | pub | GoPro360 | README | labels.json | all_jpgs.txt | buildings360-voc

Overview

This dataset contains 360-degree panoramic video frames captured with a GoPro MAX camera across multiple cities in Ohio, Kentucky, and Indiana.

Note: Original .MOV files are no longer in this directory. The extracted frames and processed data remain available.

Dataset Statistics:

360 Panorama Sequences - Browse GSxxxxxx Folders

Click any sequence to view its image gallery. From the gallery, you can access the folder directly. (52 sequences, 129,219 images total):

GS010001 GS010002 GS010003 GS010004 GS010005 GS010006 GS010008 GS010009 GS010010 GS010011 GS010014 GS020001 GS020002 GS020005 GS020006 GS020008 GS020009 GS020010 GS020011 GS020014 GS030001 GS030002 GS030005 GS030006 GS030008 GS030009 GS030010 GS030011 GS030014 GS040001 GS040006 GS040008 GS040009 GS040010 GS040011 GS040014 GS050001 GS050006 GS050008 GS050009 GS050011 GS050014 GS060001 GS060009 GS060011 GS060014 GS070001 GS070009 GS070011 GS070014 GS080011 GS080014

Directory Structure

1. GSxxxxxx Folders (52 video sequences)

Each GSxxxxxx/ folder corresponds to a single video sequence and contains:

New! Each folder now includes an html/ subdirectory with paginated image galleries. Click the green "Gallery" buttons above to browse images with navigation, thumbnails, and lightbox viewing.
Item Description
GSxxxxxx-frame-YYYYYY.jpg Original extracted panorama frames at 2fps (0.5 second intervals)
gps.csv GPS telemetry data with columns: path, ts (timestamp), lat, lon, alt (altitude), speed, speed3d, date
gps-fixed.csv Same as gps.csv but with paths pointing to fixed/ directory images and an added index column
fixed/ Panorama frames rotated to have a level horizon with north always at the same location (approximately 90 degrees in the image). Files named: GSxxxxxx-frame-YYYYYY.jpg
Includes html/ gallery subdirectory
distances-m/ Coarse depth estimate maps in TIF format (frame-YYYYYY.tif), generated using LIDAR data from USGS 3DEP
Includes html/ gallery subdirectory
html/ Auto-generated paginated image gallery (100 images per page) with thumbnails, navigation, lightbox viewing, and keyboard shortcuts
3dep/ USGS 3D Elevation Program (3DEP) LIDAR source data used for depth estimation. Contains:
  • points.laz - Point cloud data
  • dems.tar.gz - Digital Elevation Models
  • output.tin.tar.gz - Triangulated Irregular Network data
  • viewer.txt - Metadata
Note: Not all GSxxxxxx folders have this directory
buildings.geojson GIS-derived building footprints and segmentation data. Note: Not all GSxxxxxx folders have this file

2. extracted/

Contains automatically extracted building images from panoramas, organized by video sequence and extraction date.

3. annotations/

Annotation data and tools for building labeling:

New! Browse the annotated building features: View Building Annotation Galleries - Explore the LYD and Datapure subsets with interactive galleries

4. gis/

GIS building footprint data for various cities, used to automatically segment buildings in panoramas:

5. buildings/

Contains building-specific data organized by location:

6. panos/

Symbolic links to panorama images organized by city rather than by video sequence:

7. triage/

Manually reviewed extracted building images. Images that should NOT be labeled were removed before sharing with annotators.

8. pub/

Files prepared for public release or online posting.

9. GoPro360/

Originally contained proprietary .360 formatted files from GoPro camera. These files were converted to .mov format.

Status: These files may have been removed along with the MOV files.

10. Other Files

Data Processing Pipeline

  1. Video Capture: GoPro MAX 360 camera records 5.6K .360 video files
  2. Video Export: .360 files exported to .MOV format using GoPro MAX Exporter (Windows)
  3. GPS Extraction: GPS telemetry extracted from video metadata and saved to gps.csv
  4. Frame Extraction: Frames extracted at 2fps (every 0.5 seconds) as equirectangular panoramas
  5. Horizon Correction: Panoramas rotated to level horizon and normalized orientation (north at ~90 degrees), saved to fixed/
  6. LIDAR Integration: USGS 3DEP LIDAR data downloaded for locations with available coverage
  7. Depth Estimation: Coarse depth maps generated from LIDAR data, saved to distances-m/
  8. GIS Building Extraction: Building footprints from city GIS data used to segment buildings in panoramas
  9. Building Cropping: Individual buildings automatically extracted and saved to extracted/
  10. Manual Triage: Extracted buildings reviewed and filtered in triage/
  11. Annotation: Selected buildings prepared for human annotation

CSV Data Format

gps.csv

Column Description Example
path Relative path to frame image GS010001/GS010001-frame-000001.jpg
ts Timestamp in seconds from video start 0.0, 0.5, 1.0, 1.5...
lat GPS latitude in decimal degrees 39.407270
lon GPS longitude in decimal degrees -84.573313
alt Altitude in meters 184.77
speed Ground speed (m/s or km/h) 0.129
speed3d 3D speed including vertical component -0.016
date UTC timestamp 2020-09-24 21:05:33.914000+00:00

gps-fixed.csv

Same as gps.csv with these differences:

Usage Notes

Missing Files: The following directories mentioned in the original README.md are no longer present: These may be available on an external drive.
Data Availability:

Quick Reference

To find:

Documentation generated on 2025-10-17
For questions about missing data or external drive location, check with dataset maintainer.