Skip to content

Releases

LDO v0.1.0 — research preview

LDO v0.1.0 — research preview Pre-release
Pre-release

Choose a tag to compare

View all tags
@Elaybarnoam Elaybarnoam released this 16 Oct 14:05
v0.1.0
8a8ecb2

LDO v0.1.0 — research preview

LDO retrieves historical camera observations, records local correspondences and
assesses them against independent geometric evidence. This first release packages
the visual detector and historical research record for reproducible inspection.

Included

  • Installable Python 3.11 package and ldo-research commands.
  • MegaLoc + SuperPoint/LightGlue interfaces, causal replay and optional stereo/PnP
    operational verification, with external model/native prerequisites.
  • Preserved studies, mathematical definitions, saved-result tables and figures.
  • A 24-second saved-evidence walkthrough reproducible on CPU.
  • Contribution guide, starter tasks and a research roadmap.

Start from source

git clone https://github.com/Elaybarnoam/LDO.git
cd LDO
git checkout v0.1.0
uv sync --locked
uv run ldo-research doctor
uv run ldo-research smoke --output artifacts/smoke.json

Python 3.11 CPU workflows are checked on Windows and Linux in CI. The smoke
command uses synthetic adapters and must report research_result: false.
Learned inference requires external sources, checkpoints and the separate WSL/CUDA
profile described in installation. Release wheels provide
the core package and config resources; source archives provide the documentation,
examples and tests as well. Model sources and weights are not bundled.

Evidence and limits

This release introduces no new performance measurement. Confidence 0.3 remains
the baseline; 0.4 is experimental. The EfficientLoFTR candidate was not promoted.
Existing Factory/Desert and OldTownFall data have been inspected and are not fresh
tests for future selection. Reference labels are incomplete. No real-time,
real-world superiority, IMU fusion or downstream SLAM benefit is established.
The walkthrough shows archived simulated failure evidence, not live inference.

Licensing and contribution

Original code uses Apache-2.0. ORB-SLAM3 instrumentation uses GPL-3.0-or-later;
models, datasets and image derivatives retain separate upstream terms. See
licensing. Start with
contribution opportunities,
Q&A or
the roadmap.

Loading