MaineFrame Labs
// client work

Satellite image analysis

Selected client engagements in remote-sensing computer vision. Projects are described without naming clients; detailed case studies available under NDA.

// image placeholder
vessel detection
// satellite · computer vision · detection

Vessel detection in satellite imagery

End-to-end detection pipeline for identifying and localizing vessels across large satellite scenes. The core challenge wasn't the architecture — it was making a limited labeling budget go far enough, and knowing which detections to trust.

A data-centric active-learning loop prioritizes the scenes and regions most worth annotating, while uncertainty-aware scoring surfaces ambiguous detections for human review instead of silently shipping them. The result: higher effective label efficiency and a clear, honest signal on prediction confidence.

object detection remote sensing active learning uncertainty human-in-the-loop
Discuss a similar problem →
// image placeholder
biochemical analysis
// satellite · analysis · regression

Biochemical analysis from satellite imagery

Models that infer biochemical indicators from satellite data. The harder part is reliability: satellite inputs shift with season, sensor, and geography, and a confident-but-wrong estimate is worse than an honest "uncertain".

Work focused on distribution-shift awareness and confidence calibration — emphasizing when a prediction is reliable enough to act on and when it should be flagged. Evaluation was built around decision quality, not just regression error.

regression calibration distribution shift uncertainty
Discuss a similar problem →
// note  Client names and dataset specifics are withheld for confidentiality. If a case study here looks relevant to your problem, reach out and we can talk specifics under NDA.