Quick Start Guide

Quick Start Guide#

Get started with Clay Foundation Model in 5 minutes!

1. Install#

pip install git+https://github.com/Clay-foundation/model.git

2. Download Weights#

wget https://huggingface.co/made-with-clay/Clay/resolve/main/v1.5/clay-v1.5.ckpt

3. Generate Embeddings#

import yaml
import torch
from claymodel.module import ClayMAEModule

# Load model
model = ClayMAEModule.load_from_checkpoint("clay-v1.5.ckpt")
model.eval()

# Load sensor metadata
with open("configs/metadata.yaml", "r") as f:
    metadata = yaml.safe_load(f)

# Prepare Sentinel-2 data
sensor = "sentinel-2-l2a"
chips = torch.randn(1, 10, 256, 256)  # [batch, bands, height, width]

# Get wavelengths from metadata (convert μm to nm)
wavelengths = []
for band in metadata[sensor]["band_order"]:
    wavelengths.append(metadata[sensor]["bands"]["wavelength"][band] * 1000)
wavelengths = torch.tensor([wavelengths], dtype=torch.float32)

timestamps = torch.zeros(1, 4)  # [week, hour, lat, lon] - can be zeros

# Generate embeddings
with torch.no_grad():
    embeddings = model.encoder(chips, timestamps, wavelengths)

print(f"Embeddings shape: {embeddings.shape}")  # [1, 1024]

4. Next Steps#

Supported Sensors#

Clay v1.5 is sensor-agnostic and works with any satellite instrument! Currently supported sensors include:

Sensor

Bands

Resolution

Description

Sentinel-2 L2A

10

10m

Optical multispectral

Landsat C2 L1/L2

6

30m

Optical multispectral

NAIP

4

1m

Aerial RGB + NIR

LINZ

3

0.5m

Aerial RGB

Sentinel-1

2

10m

SAR (VV, VH)

MODIS

7

500m

Global surface reflectance

 
your-satellite:
  band_order: [blue, green, red, nir]    # Your band names
  gsd: 10.0                              # Resolution in meters
  bands:
    wavelength: {blue: 0.485, green: 0.560, red: 0.660, nir: 0.835}  # μm
    mean: {blue: 1200, green: 1400, red: 1600, nir: 2800}           # Normalization
    std: {blue: 400, green: 450, red: 500, nir: 650}                # Statistics

Need Help?#

Contents