Classify the sky — one frame at a time.
SkyClf is a lightweight, self-hosted system for classifying all-sky camera images
and exposing a simple, reliable “current sky state” API for observatories.
SkyClf continuously fetches the latest all-sky image, lets you label frames and serves the result via a clean HTTP API. It is designed for local training, because no two skies — or cameras — are the same.
Supported States:
clear · light_clouds · heavy_clouds · precipitation · unknown
| Project | Stack | Description | Last Update |
|---|---|---|---|
| SkyClf | Go, Docker | Core Application. Handles fetching, labeling, API, and model inference. |  |
| SkyClf-Trainer | Python | Training Pipeline. Reads local labels → trains a model → exports versioned models. |  |
All-sky setups vary wildly regarding optics, sensors, light pollution, and horizon obstructions. SkyClf embraces this by making user-trained models first-class, not an afterthought.
- API Entry Point:
GET /api/latest - Contribute: PRs are welcome! New classes or tasks should include clear labeling rules.
Clear nights are rare — so we spend a lot of time training on clouds. ☁️