I live in the Arctic Circle, where the Northern Lights are frequently visible. I set up a 24/7 live stream of the sky, both as an early warning system for when the aurora is active and to share the view with everyone.
As part of this project, I wanted to hook a computer vision model to the video feed to automatically detect auroral activity and send notifications. I quickly ran into two problems:
1. There were no pre-trained models for this, which meant I had to train my own. 2. Training a model requires a large, well-organized dataset. Getting the dataset right involves frequent sorting, tweaking, and updating thousands of images.
I tried to manage my images with traditional file explorers, Darktable, and many other tools, but none of them felt efficient enough for the task. This led me down a bit of a yak-shaving journey, and Picsort was born.
Picsort is a keyboard-first desktop app (with Vim-like HJKL navigation) designed to do one thing: rapidly sort large batches of images into folders.
It's cross-platform (Linux, Windows, macOS), non-destructive, and built in Go. It generates a thumbnail/preview cache on first load so that subsequent navigation is fast. While I built it for my CV dataset, it's versatile enough for any large-scale photo organization task.
I'd love to get your feedback.
GitHub (MIT License): https://github.com/coolapso/picsort Website (with demo & downloads): https://picsort.coolapso.sh
And if you're curious, here's the Northern Lights stream that started it all: https://youtube.com/@thearcticskies