The Pipeline

How Kestrel Sees Your Photos.

Project Kestrel doesn't just "guess" which photos are good. It follows a multi-stage machine learning pipeline to decompose your photography into objective data.

Phase 1

Intelligent Scene Grouping

Kestrel compares your images with each other to identify high-speed bursts and groups them into scenes. By grouping these into scenes, you can compare nearly identical frames against each other using Kestrel's quality algorithms.

Timeline of photos grouping into scenes

Phase 2

Subject Detection & Masking

Using a specialized object detection model, Kestrel identifies birds within the frame. It then creates precise masks around the subjects, ensuring that background detail never influences the quality score.

Masking process showing original bird to masked crop

Phase 3

Objective Quality Analysis

Kestrel's machine learning model is trained to take into account noise, motion blur, and sharpness to create a normalized quality score that can accurately differentiate between each frame in a scene.

Sharpness heatmap highlighting key detail areas on a bird

Phase 4

Species Classification (Beta)

Kestrel automatically tags your photos by species and family. This powerful metadata allows you to search through tens of thousands of photos instantly to re-discover your favorite birding memories.

Kestrel identifying bird species and families automatically

Design Philosophy

Computers can never replace artistic vision.

Project Kestrel is built on the belief that AI should be a cofactor, not a replacement. Kestrel handles the "boring" parts—calculating sharpness and grouping bursts—so you can focus on the artistic decisions: composition, lighting, and story.

100% Local. 100% Yours. All models run on your CPU/GPU. No images are ever sent to a server.