Satellite Cloud Detection (with Triton Fallback)

Upload separate JP2 files for Red (e.g., B04), Green (e.g., B03), Blue (e.g., B02), and NIR (e.g., B8A) channels. The application will first attempt to use a remote Triton Inference Server. If the server is unavailable or inference fails, it will fall back to using the local OmniCloudMask model.

Pixel Classification:

Clear (Green)
Thick Cloud (Red)
Thin Cloud (Yellow)
Cloud Shadow (Blue)

The model works best with imagery at 10-50m resolution.

Input Bands (JP2)

Red Channel (e.g., B04)

Green Channel (e.g., B03)

Blue Channel (e.g., B02)

NIR Channel (e.g., B8A)

Local Model Parameters (Used for Fallback)

Batch Size

Memory usage/speed for local model

1 32

Patch Size

Patch size for local model processing

256 2048

Patch Overlap

Overlap for local model processing

64 512

Results

Original RGB Image (Approx. True Color)

Cloud Detection Visualization (Mask Overlay)

Processing Status & Statistics

Examples

Red Channel (e.g., B04)	Green Channel (e.g., B03)	Blue Channel (e.g., B02)	NIR Channel (e.g., B8A)	Batch Size	Patch Size	Patch Overlap