OLF - Open Labeling Format

Brief

The Open Labeling Format is a specification aimed at labeling (not only) automotive sensor data like below:

Details

For a complete description of the OLF schema please read the white-paper. Please also have a look at the schema file itself.

Overview

.
├── Documentation
│   └── OLF_WhitePaper.pdf                    # White-paper and in-depth description of the format
├── Documents
│   └── example.olf                           # Sample OLF file (dummy)
├── License.md                                # License file
├── OpenLabelingFormat.lxsopt                 # Project file(s) for Liquid Studio
├── OpenLabelingFormat.lxsproj                # Project file(s) for Liquid Studio
├── README.md                                 # This file
├── Resources
│   ├── labeling.jpg                          # Sample screen of labeling process
│   ├── hagl_logo.jpg                         # HELLA Aglaia logo
│   └── olf_logo.jpg                          # OLF logo
├── Schemas
│   └── schema.olf.json                       # Actual OLF specification as JSON schema
└── Tools
    ├── Converter
    │   ├── COCO
    │   │   └── coco2olf.py                   # Converter from MS COCO to OLF
    │   └── TFRecord
    │       └── OLFDetection2DTFRecord.py     # Converter from OLF to TFRecord
    ├── Loader
    │   ├── OLFDetection2D.py                 # Base class for extracting 2D detections from an OLF file
    │   └── PyTorch
    │       └── OLFDetection2DDataset.py      # Detection dataset class for PyTorch based on OLFDetection2D.py
    └── Notebooks
        ├── Demo_OLFDetection_PyTorch.ipynb   # Demo notebook showing how to use OLFDetection2DDataset.py
        └── Demo_OLFDetection_TFRecord.ipynb  # Demo notebook showing how to use OLFDetection2DTFRecord.py

Requirements

The OLF schema has been designed using Liquid Studio 2019 - JSON Editor Edition 17.1.5.9520 offered by Liquid Technologies. It is based on JSON schema draft-07.

Python scripts require the following package versions:

Package	Version
jsonschema	3.1.1
object-detection	0.0.3
pandas	0.25.1
python	3.6.9
scikit-image	0.15.0
tensorflow	2.0.0
torch	1.3.0

Quick-Start

To convert detections from MS COCO into OLF, check-out this script.

For a demo on how to use an OLF file as a PyTorch detection dataset, check-out this notebook.

For a demo on how to convert an OLF file into a detection TFRecord, checkout this notebook.

License

The OLF standard, the schema file(s) and the documentation file(s) are licensed under a
Creative Commons Attribution-ShareAlike 4.0 International License.

For all other parts of the repository this license applies.

Maintainer & Copyright

This repository is currently maintained by HELLA Aglaia Mobile Vision GmbH.

Copyright (c) 2020, HELLA Aglaia Mobile Vision GmbH