Open-core · Active research

More frames per pass. No extra satellites.

Coding Pirates synthesizes intermediate satellite frames using deep learning optical flow — turning a 30-minute INSAT-3DS cadence into 7.5-minute near-real-time coverage for disaster monitoring, without additional hardware.

≥ 0.87
SSIM quality score
vs ground truth
≥ 34 dB
PSNR
peak signal-to-noise
Temporal resolution
30 min → 7.5 min
4
Satellite sources
INSAT-3DS/R · GOES-19 · Himawari
MIT
License
Open-core model weights
01How it works

Optical flow interpolation
at satellite scale.

We ingest raw thermal infrared data from geostationary satellites, extract motion vectors with RIFE-based optical flow, and synthesize photometrically valid intermediate frames that have never been captured by the sensor.

1
Ingest
ingestion
Read raw .nc/.h5 TIR data from INSAT-3DS, INSAT-3DR, GOES-19, or Himawari. Normalize, reproject, and quality-check each frame pair.
2
Interpolate
inference
RIFE / Super SloMo deep learning backbone computes bidirectional optical flow and warps both frames to synthesize one or three intermediate frames.
3
Deliver
output
Output GeoTIFF + NetCDF with SSIM, PSNR, FSIM metrics reported per frame. Push to the web dashboard for playback and ground-truth comparison.
02Capabilities
CAP-01
Multi-satellite support
Native support for INSAT-3DS/3DR (ISRO), GOES-19 (NOAA), and Himawari (JMA). Pluggable reader architecture for additional GEO sources.
CAP-02
Quality-validated output
Every synthesized frame ships with SSIM, PSNR, MSE, and FSIM metrics computed against withheld ground truth. Drift and artifact reports included.
CAP-03
Web dashboard
Side-by-side animation playback of interpolated vs. ground truth sequences. Heatmap overlays, per-frame quality charts, and export controls.
CAP-04
Open-core model
Model weights, training code, and the CLI pipeline are MIT/Apache licensed and published on GitHub. Enterprise users access managed inference and SLA-backed APIs.
CAP-05
Managed inference API
REST and WebSocket endpoints for production integration. Pay-per-call or monthly subscription. White-labeled dashboard available for agency deployments.
CAP-06
On-premise deployment
Professional services for fine-tuning on proprietary agency datasets and deploying the inference stack inside air-gapped or restricted network environments.
03Use cases

Built for the agencies
that cannot wait.

Dynamic weather events evolve faster than the 30-minute revisit window most geostationary satellites deliver. We close that gap with synthetic frames — validated against real high-temporal data — so meteorologists have more decision points during critical hours.

Cyclone tracking
High-temporal imagery shows landfall timing and intensity changes more precisely. Emergency managers get more decision points before evacuation orders.
Flash flood monitoring
Mesoscale convective systems develop within minutes. Dense intermediate frames help nowcast systems issue warnings ahead of onset.
Wildfire spread
Infrared frame interpolation tracks fire perimeter changes between scans, improving resource dispatch accuracy.
Climate research
Higher-cadence sequences give researchers more data density for studying cloud dynamics, ocean heat uptake, and atmospheric circulation.
Enterprise GIS products
Agriculture, insurance, and logistics companies consuming satellite-derived products benefit directly from higher temporal granularity in their risk models.
04Open-core model

The model is open.
The platform scales.

Model weights, training scripts, and the CLI pipeline are published under MIT/Apache license. Agencies and researchers can validate the approach independently. Enterprise tiers add managed hosting, SLA-backed inference endpoints, and dedicated support.

Open Source
  • Model weights
  • Training code
  • CLI pipeline
  • MIT/Apache license
Enterprise
  • Managed inference API
  • SLA-backed endpoints
  • White-label dashboard
  • Fine-tuning support

Get early access to the platform.

We work with national meteorological agencies and research institutions.

Request access