Prithvi Models Usage¶

This notebook demonstrates how to use the Prithvi models for satellite image reconstruction, with simple examples covering all six Prithvi foundation MAE models.

In [ ]:

# %pip install geoai-py

# %pip install geoai-py

Import Library¶

In [ ]:

import geoai
import matplotlib.pyplot as plt

import geoai import matplotlib.pyplot as plt

List Available Models¶

In [ ]:

models = geoai.get_available_prithvi_models()
models

models = geoai.get_available_prithvi_models() models

Load Models¶

In [ ]:

processor = geoai.load_prithvi_model("Prithvi-EO-2.0-tiny-TL")

# processor_tiny = geoai.load_prithvi_model("Prithvi-EO-2.0-tiny-TL")
# processor_100m = geoai.load_prithvi_model("Prithvi-EO-2.0-100M-TL")
# processor_300m = geoai.load_prithvi_model("Prithvi-EO-2.0-300M")
# processor_300m_tl = geoai.load_prithvi_model("Prithvi-EO-2.0-300M-TL")
# processor_600m = geoai.load_prithvi_model("Prithvi-EO-2.0-600M")
# processor_600m_tl = geoai.load_prithvi_model("Prithvi-EO-2.0-600M-TL")

processor = geoai.load_prithvi_model("Prithvi-EO-2.0-tiny-TL") # processor_tiny = geoai.load_prithvi_model("Prithvi-EO-2.0-tiny-TL") # processor_100m = geoai.load_prithvi_model("Prithvi-EO-2.0-100M-TL") # processor_300m = geoai.load_prithvi_model("Prithvi-EO-2.0-300M") # processor_300m_tl = geoai.load_prithvi_model("Prithvi-EO-2.0-300M-TL") # processor_600m = geoai.load_prithvi_model("Prithvi-EO-2.0-600M") # processor_600m_tl = geoai.load_prithvi_model("Prithvi-EO-2.0-600M-TL")

Download Sample Data¶

In [ ]:

url = "https://huggingface.co/datasets/giswqs/geospatial/resolve/main/Mexico_HLS_samples.zip"
geoai.download_file(url)

url = "https://huggingface.co/datasets/giswqs/geospatial/resolve/main/Mexico_HLS_samples.zip" geoai.download_file(url)

Define Input Files¶

Specify the paths to the satellite images for processing.

In [ ]:

file_paths = [
    "Mexico_HLS_samples/data/Mexico_HLS.S30.T13REM.2018026T173609.v2.0_cropped.tif",
    "Mexico_HLS_samples/data/Mexico_HLS.S30.T13REM.2018106T172859.v2.0_cropped.tif",
    "Mexico_HLS_samples/data/Mexico_HLS.S30.T13REM.2018201T172901.v2.0_cropped.tif",
    "Mexico_HLS_samples/data/Mexico_HLS.S30.T13REM.2018266T173029.v2.0_cropped.tif",
]

file_paths = [ "Mexico_HLS_samples/data/Mexico_HLS.S30.T13REM.2018026T173609.v2.0_cropped.tif", "Mexico_HLS_samples/data/Mexico_HLS.S30.T13REM.2018106T172859.v2.0_cropped.tif", "Mexico_HLS_samples/data/Mexico_HLS.S30.T13REM.2018201T172901.v2.0_cropped.tif", "Mexico_HLS_samples/data/Mexico_HLS.S30.T13REM.2018266T173029.v2.0_cropped.tif", ]

Simple One-Line Inference¶

Run inference using the simplified API.

In [ ]:

geoai.prithvi_inference(
    file_paths=file_paths, model_name="Prithvi-EO-2.0-tiny-TL", output_dir="results"
)

geoai.prithvi_inference( file_paths=file_paths, model_name="Prithvi-EO-2.0-tiny-TL", output_dir="results" )

Advanced Processing with Processor¶

Load the model and process files with custom parameters.

In [ ]:

processor = geoai.load_prithvi_model()
output_dir = "results"
processor.process_files(file_paths, output_dir, mask_ratio=0.75)

processor = geoai.load_prithvi_model() output_dir = "results" processor.process_files(file_paths, output_dir, mask_ratio=0.75)

Visualize Results¶

Process images and extract tensors for visualization.

In [ ]:

input_tensor, rec_tensor, mask_tensor = processor.process_images(
    file_paths=file_paths, mask_ratio=0.75
)

input_tensor, rec_tensor, mask_tensor = processor.process_images( file_paths=file_paths, mask_ratio=0.75 )

In [ ]:

original_rgb, masked_rgb, reconstructed_rgb = processor.visualize_rgb(
    input_tensor, rec_tensor, mask_tensor
)

original_rgb, masked_rgb, reconstructed_rgb = processor.visualize_rgb( input_tensor, rec_tensor, mask_tensor )

Plot Comparison¶

Display the original, masked, and reconstructed images side by side.

In [ ]:

num_timestamps = len(original_rgb)
fig, axes = plt.subplots(num_timestamps, 3, figsize=(12, 4 * num_timestamps))

if num_timestamps == 1:
    axes = axes.reshape(1, -1)

column_titles = ["Original", "Masked (Visible Patches)", "Reconstructed"]

for t in range(num_timestamps):
    axes[t, 0].imshow(original_rgb[t])
    axes[t, 0].set_title(f"T{t+1}: {column_titles[0]}")
    axes[t, 0].axis("off")

    axes[t, 1].imshow(masked_rgb[t])
    axes[t, 1].set_title(f"T{t+1}: {column_titles[1]}")
    axes[t, 1].axis("off")

    axes[t, 2].imshow(reconstructed_rgb[t])
    axes[t, 2].set_title(f"T{t+1}: {column_titles[2]}")
    axes[t, 2].axis("off")

plt.tight_layout()
plt.show()

num_timestamps = len(original_rgb) fig, axes = plt.subplots(num_timestamps, 3, figsize=(12, 4 * num_timestamps)) if num_timestamps == 1: axes = axes.reshape(1, -1) column_titles = ["Original", "Masked (Visible Patches)", "Reconstructed"] for t in range(num_timestamps): axes[t, 0].imshow(original_rgb[t]) axes[t, 0].set_title(f"T{t+1}: {column_titles[0]}") axes[t, 0].axis("off") axes[t, 1].imshow(masked_rgb[t]) axes[t, 1].set_title(f"T{t+1}: {column_titles[1]}") axes[t, 1].axis("off") axes[t, 2].imshow(reconstructed_rgb[t]) axes[t, 2].set_title(f"T{t+1}: {column_titles[2]}") axes[t, 2].axis("off") plt.tight_layout() plt.show()