Series: Oriented-Det
Technical notes on oriented-det: announcements, releases, training, evaluation, and parity work against research frameworks.
Start here
- Oriented-det is coming — motivation and design goals
- Oriented-Det v0.1.0 is out — install, docs, and getting started
- Oriented R-CNN detections for the 15 DOTA classes — qualitative gallery on the DOTA taxonomy
- Oriented object detection on macOS, in pure Python — hands-on inference with
odet image-demoon Apple Silicon - Zero-shot ships on Sentinel-2 — public checkpoint on a Copernicus tile
- Sliding-window inference on large aerial tiles — pad, tile, merge NMS with the Oriented R-CNN 1× Hub weight
- Rotated Faster R-CNN on DOTA without custom CUDA — ProbIoU, sampled rIoU, 74.42% official Task 1 vs MMRotate 73.40
- Oriented-Det v0.1.1 — ProbIoU packaged, MMRotate parity, harbor-scene demo
- Oriented-Det v0.2.0 — Rotated FCOS, decoded rIoU, four-family 1×/3× official Task 1 zoo
- Rotated FCOS vs Oriented R-CNN on macOS — Apple Silicon MPS latency, 1× L4 training wall, side-by-side demos
- A static demo of three oriented detectors — Oriented R-CNN 3×, Rotated Faster R-CNN 3×, and FCOS 3× on six optical scenes, in parity with MMRotate
- Apache 2.0 covers oriented-det. It does not cover DOTA or HRSC. — sovereignty of the stack versus research datasets; train on your own licensed imagery
- HRSC2016 recipes, trains, and results — native ship loader, three 3× Hub weights, held-out test 90.41% / 88.77% / 88.34%
- Lessons learned on DOTA — official Task 1, the leaky eval-val trap, and why the last mile was the box loss
- Oriented-Det v0.3.1 — four dataset loaders, HRSC Hub 3×, RetinaNet OBB, ONNX export
- FAIR1M fine-grained detection — 37 classes, convert/tile, Faster R-CNN 1× tiled-val 36.70%
- SSDD SAR ship finetune — optical DOTA → SAR, held-out test 90.34%
- HRSID SAR ship benchmark — larger SAR set, 78.55% rotated AP50 vs Wei HBB
- Deploy in Docker — Sanic Tile Geo Process, GeoJSON out, PyTorch in CUDA
- ONNX export without PyTorch — pre-NMS ONNX + ORT consumer stack
- Does MMRotate Faster R-CNN also mess up the bus lot? — official 1× Rotated Faster R-CNN is messy on the same DOTA bus tile as OrientedDet FRCNN (architecture, rare ~45° clutter); Oriented R-CNN is clean but heavy, which pushes toward FCOS
- Which oriented detector should you train? — Oriented R-CNN when the box must be tight (AP75); Faster R-CNN or FCOS when recall matters more, with FCOS for dense ~45° objects
Links
- GitHub · PyPI · Documentation
-
FAIR1M in oriented-det — 37 classes, and why 36.70% is not a failed train
Native FAIR1M support in oriented-det v0.3: convert + tile, finetune DOTA 1× Faster R-CNN, tiled-val 36.70% mAP50. The bottleneck is class ID, not boxes. No Hub zoo — CC BY-NC-SA dump.
-
Oriented-Det v0.3.1 — four datasets, ONNX, and RetinaNet OBB
Oriented-det v0.3.1 is on PyPI — native HRSC2016, FAIR1M, SSDD, and HRSID loaders, HRSC Hub 3× zoo, DOTA advertised from 1×, RetinaNet Hub OBB, and ONNX export.
-
Lessons learned on DOTA: official Task 1, leaky val, and the last mile
What I would keep from four months on DOTA: quote official Task 1, never quote leaky eval-val, and treat MMRotate as a reference in the same band — the last mile was the box loss, not a new head.
-
HRSC2016 in oriented-det — recipes, trains, and a held-out 90.41%
Native HRSC2016 ship detection in oriented-det: ImageSets trainval in, held-out test out, three 3× Hub weights. Oriented R-CNN 90.41%, Faster R-CNN 88.77%, FCOS 88.34% mAP50 on NVIDIA L4.
-
Apache 2.0 covers oriented-det. It does not cover DOTA or HRSC.
The oriented-det framework is Apache 2.0. DOTA and HRSC are research datasets with no commercial grant. Production detectors are trained on your own licensed imagery — this is not legal advice.
-
A static demo of three oriented detectors on optical satellite imagery
A browser demo of Oriented-Det’s three 3× DOTA checkpoints — Oriented R-CNN, Rotated Faster R-CNN, and Rotated FCOS — on six optical satellite scenes, with a side by side comparison that lands in parity with state of the art frameworks.
-
Rotated FCOS vs Oriented R-CNN on macOS
Hands-on Apple Silicon comparison of Hub 1× DOTA checkpoints — Rotated FCOS (73.07% official Task 1) vs Oriented R-CNN (76.73%) — MPS latency, 1× training wall on NVIDIA L4, score thresholds, and side-by-side detections.
-
Oriented-Det v0.2.0 — Rotated FCOS, decoded rIoU, and a four-family zoo
Oriented-det v0.2.0 is on PyPI — Rotated FCOS joins the zoo as the balanced one-stage detector, with a decoded rIoU 1× Hub checkpoint at 73.07% official DOTA Task 1, and the same Apache 2.0 stack.
-
Oriented-Det v0.1.1 — ProbIoU, MMRotate parity, and the updated zoo
Oriented-det v0.1.1 is on PyPI — ProbIoU ROI regression, MMRotate-aligned training fixes, a DOTA le90 zoo on official Task 1 led by Oriented R-CNN 1× at 76.73%, and a hands-on harbor-scene demo of the Faster R-CNN throughput pick.
-
Rotated Faster R-CNN on DOTA without custom CUDA: sampled rIoU, ProbIoU, and a 74.42% Task 1 checkpoint
Why OrientedDet avoids MMRotate's exact CUDA IoU kernels, how ProbIoU trains oriented boxes in pure PyTorch, and why the 1× Rotated Faster R-CNN Hub weight beats MMRotate on official DOTA Task 1.
-
Sliding-window inference on images larger than the DOTA canvas
How odet image-demo tiles images larger than 1024×1024, merges overlapping windows, and filters classes — using the Oriented R-CNN 1× Hub checkpoint retrained after a diagonal-flip bug.
-
Zero-shot Ship detection on a Copernicus Sentinel-2 tile with Oriented R-CNN
A practical zero-shot experiment with a DOTA-pretrained Oriented R-CNN model on a 10 m Copernicus Sentinel-2 tile, including zoomed sliding-window inference and ship-only filtering.
-
Oriented object detection on macOS, in pure Python
Run Oriented R-CNN on Apple Silicon with pip, uv, and the odet CLI — no CUDA toolchain, from demo.jpg to result.jpg in minutes.
-
Oriented R-CNN detections for the 15 DOTA classes
A visual tour of Oriented R-CNN detections across the 15 original DOTA v1.0 aerial object classes.
-
Oriented-Det v0.1.0 is out — sovereign oriented object detection for EO
Oriented-det v0.1.0 is on PyPI under Apache 2.0 — a lightweight PyTorch library for rotated object detection in aerial and satellite imagery, with DOTA baselines and pretrained weights.
-
Oriented-det is coming: sovereign oriented detection for EO
A sovereign, Apache-licensed oriented detection stack for EO, targeting an official release in June 2026.