Spaces:
Sleeping
Sleeping
| title: OlmoEarth Vision Explorer | |
| emoji: 🛰️ | |
| colorFrom: blue | |
| colorTo: green | |
| sdk: gradio | |
| sdk_version: 6.10.0 | |
| app_file: app.py | |
| short_description: OlmoEarth satellite embedding visualization demo | |
| startup_duration_timeout: 1h | |
| # OlmoEarth Vision Explorer | |
| Interactive demo of the [OlmoEarth-v1-Base](https://huggingface.co/allenai/OlmoEarth-v1-Base) foundation model for Earth observation. | |
| ## Features | |
| - Real satellite imagery gallery (10 land cover classes from EuroSAT / Sentinel-2) | |
| - Pre-computed model embeddings load instantly | |
| - Upload your own satellite imagery for live inference | |
| - PCA projection of patch embeddings to RGB | |
| - Per-patch similarity heatmap | |
| - Split-screen comparison view | |
| - Model info overlay | |
| ## Example Imagery | |
| All 10 cached examples are real Sentinel-2 satellite images from the [EuroSAT](https://huggingface.co/datasets/blanchon/EuroSAT) dataset: | |
| - Annual Crop, Forest, Herbaceous Vegetation, Highway, Industrial Buildings | |
| - Pasture, Permanent Crop, Residential Buildings, River, Sea & Lake | |
| Each example includes pre-computed PCA visualization, similarity heatmap, and embedding statistics — all available instantly without running inference. | |
| ## Technical | |
| - ViT-Base encoder (89M params, 768-dim embeddings) | |
| - 56×56 patch grid (3,136 patches per image) | |
| - CPU inference | |
| - `gr.Server` custom fullscreen HTML/CSS/JS frontend | |
| - Dark, minimalist, mobile-responsive design | |