license: mit
library_name: libreyolo
pipeline_tag: object-detection
tags:
- object-detection
- oriented-object-detection
- gtr
- pytorch
- libreyolo
- dota
datasets:
- DOTA
LibreGTRs-obb
GTR-S oriented-box weights for DOTA v1.0, converted for LibreYOLO.
GTR support is being prepared for LibreYOLO v1.6.0. Earlier PyPI releases may not include this model family.
Usage
With a LibreYOLO version that includes GTR OBB:
from libreyolo import LibreYOLO
model = LibreYOLO("LibreGTRs-obb.pt")
result = model.predict("aerial.jpg")
result.obb.xywhr # (cx, cy, w, h, theta) in pixels, theta in [0, pi)
The input is a fixed 1024 by 1024 pixel square. Images are resized to fit and padded at the bottom and right. Class ids follow the upstream DOTA v1.0 order: plane, baseball-diamond, bridge, ground-track-field, small-vehicle, large-vehicle, ship, tennis-court, basketball-court, storage-tank, soccer-ball-field, roundabout, harbor, swimming-pool, helicopter. This checkpoint is inference-only in LibreYOLO.
Source
Official GTR implementation,
source revision 782e737efe2e6437ac537fbdcee089673d3376c1.
Published checkpoint,
weight repository revision 9fc62c8c2b2c976835d0f1c1ffc544dbc0f9e29f.
Copyright (c) 2026 Intellindust-AI-Lab. The source code is MIT licensed and the
publisher's weight repository explicitly declares MIT. The publisher reports
80.0 AP50 on the DOTA-v1.0 test set; LibreYOLO has not reproduced it.
Modifications
Selected the EMA state dict and added LibreYOLO schema v1.0 metadata.
Learned parameters and state-dict keys are unchanged. Training/optimizer state
was removed. Conversion uses weights/convert_gtr_weights.py in the
LibreYOLO source repository.
Validation
Checkpoint schema, exact tensor preservation and strict loading were checked for this artifact. On CPU its logits and rotated boxes match the pinned upstream graph exactly, and decoded boxes match upstream's post-processor. DOTA benchmark accuracy and GPU latency were not independently reproduced.