// the take
A curated link collection — not a codebase — indexing deep learning techniques for satellite and aerial imagery. It's a single massive README organized by task type: classification, segmentation, object detection, change detection, SAR, and more. Aimed at researchers and ML engineers working in the geospatial domain who need a starting point for literature and implementations.
Coverage is genuinely broad and kept reasonably current — the July 2026 push date suggests active maintenance, and the inclusion of recent work like foundation models, SAM adaptations, and LoRA fine-tuning of Prithvi/TerraMind shows it tracks the field. The task-based organization (segmentation split by water, fire, crops, landslides, etc.) makes it faster to navigate than a flat list. It links directly to GitHub repos rather than just papers, which is more useful for practitioners. The companion website suggests some effort to present this as more than a raw dump.
It's a link list, not a framework — there's no code, no benchmarks, no way to know which of the 400+ linked repos actually work or are maintained. Quality varies wildly across the linked projects and there's no signal for it. The README is so long it's effectively unsearchable without Ctrl+F, and the 'how to use' instruction literally says to use Ctrl+F. No opinionated guidance on which approaches work best for which data types or scales — a newcomer gets quantity, not direction.
We read the README, the directory tree and a key source file in July 2026. Nobody pays to be listed. How we judge →