SnapNature: Identify Nature with Open-Source AI — Touch Grass Challenge

Description and use

SnapNature is a PWA that enables quick, privacy‑first identification of plants and animals using open HuggingFace models, functioning offline.

Problem

Users want a mobile-friendly way to identify plants, animals, and natural scenes without relying on proprietary APIs or internet connectivity.

Outcome

SnapNature enables instant, privacy‑first nature identification on any device, offline, with open‑weight models and easy model swapping.

Workflow

• Take a photo of a natural object
• Upload or capture image in the web interface
• Send image to HuggingFace ViT model for inference
• Receive top‑5 predictions with confidence scores
• Display educational facts and store observation in local journal
• Optionally swap the model for fine‑tuned regional models

Evidence

SnapNature uses a pre-trained ResNet50 model (ImageNet) to classify images.

Source

Take a photo of any plant, animal, or nature scene

Source

Works offline via Service Worker — go to a trail, no signal needed

Source