DL4EO — Deep Learning for Earth Observation

DL4EO builds sovereign oriented detection for satellite and aerial imagery: open-source tooling, transparent practices, and shared knowledge for production-grade rotated object detection.

Oriented bounding boxes are the core representation, not a side feature. They capture heading, footprint, and dense-scene geometry that axis-aligned detectors miss — which matters for ships, aircraft, vehicles, and crowded scenes such as harbors and airports.

OrientedDet — Open-source PyTorch framework for oriented and rotated object detection in aerial and satellite images

We maintain oriented-det, an Apache-licensed PyTorch framework for rotated object detection. You can run it on-prem, in a private cloud, or in a regulated environment — no hosted-inference requirement, no platform lock-in.

Overview: oriented detection in aerial and satellite images · Code: GitHub · Docs: dl4eo.github.io/oriented-det

Getting started

pip install oriented-det

Pretrained DOTA checkpoints live on this Hub:

odet pretrained list
odet pretrained download oriented_rcnn_dota_le90_1x

Baselines include Oriented R-CNN, Rotated Faster R-CNN, Rotated RetinaNet, and Rotated FCOS (ResNet + FPN). Docs: dl4eo.github.io/oriented-det.

On this Hub

• Models — dl4eo/oriented-det-pretrained: checkpoints for fine-tuning and inference
• Spaces
  • Object Detection Optical Satellite — oriented-det demo on real EO scenes
  • Open Detection Optical Satellite — detect objects from a text prompt
  • Ship Detection · Aircraft Detection

Work with us

Training, consulting, and project delivery around oriented-det: dl4eo.com · jeff@dl4eo.com