SmartWilds¶
Integration, alignment, and release code for a multi-modal field sensing dataset
SmartWilds is a multi-modal sensing dataset recorded with camera traps, acoustic recorders, drones, and visitor photographs at The Wilds (Cumberland, Ohio) in summer 2025. This repository is the home of the SmartWilds code and documentation. The first release is the metadata: Darwin Core tables for all four modalities on one Event hierarchy, with corrected timestamps and coordinates so they can be joined in space and time, the record of every defect found and repaired, and the code that did it. Raw images, recordings, and labels will be released soon and documented here as well.
Figure 1. Sensor deployment across The Wilds: camera traps with viewing angles, Song Meter recorders, citizen-science photo locations, and drone flight tracks. Satellite imagery © Google.
Figure 2. The same three managed focal species (Sichuan takin, Milu deer, Przewalski's horse) observed by camera trap, drone, and citizen-science photography.
What is released¶
The Darwin Core Event and Occurrence tables, the dataset cards, the supplemental material and defect inventory, the field notes and campaign reports, and the code that produced and validated all of it. Raw images and labels will be released soon. The 2025 pilot media are already in the SmartWilds collection.
| Part | Where |
|---|---|
| How to join the modalities | Joining in space and time |
| Darwin Core tables | ABC-Center/smartwilds_metadata on Hugging Face (guide) |
| Dataset cards for the five SmartWilds datasets | metadata/ |
| Supplemental material, defect inventory, camera-trap release note | docs/ |
| Data summaries and campaign data reports | field_campaigns/ |
| Generation, validation, georeferencing, screening scripts | scripts/ |
| Applications: cross-modal ecology, CV benchmarking, systems planning, biodiversity monitoring, deployment design, integration methods | Applications |
| How the agents were used, and how to use them on your data | Agents |
Getting started¶
git clone https://github.com/Imageomics/smartwilds
cd smartwilds
pip install -r requirements.txt
huggingface-cli download --repo-type dataset ABC-Center/smartwilds_metadata --local-dir smartwilds_metadata
python examples/quickstart.py smartwilds_metadata/smartwilds-camera-traps smartwilds_metadata/smartwilds-bioacoustics
The quickstart loads the tables, applies a verification floor, joins Occurrences to their Events, and summarizes them by deployment and by class. See Using the data correctly before computing anything from the Occurrence tables.
Papers¶
Several papers build on SmartWilds; each describes one part of the project.
- SmartWilds: Integrating, Aligning, and Documenting Multi-Modal Field Sensing Data (under review): the metadata and data engineering documented on this site. The supplemental material is here.
- SmartWilds: Multimodal Wildlife Monitoring Dataset (2025): the four-day pilot deployment and its first data release.
- A Blueprint for Cross-Modal Coordination at the Far Edge (to appear): planning cross-modal sensing systems from the metadata.
- Cross-Modal Corroboration for Annotation-Free Wildlife Monitoring (2026): annotation-free verification of Milu deer activity.
- Resolve, Clear, or Defer: Training-Free Active Prototype Triage for Ecological Detection Pipelines (2026): the semi-automated labeling method behind the camera-trap labels, evaluated on SmartWilds and other datasets.
Citation¶
See CITATION.cff.
Please cite the paper and each dataset you use.