Metadata-Version: 2.5
Name: thirdeyelabs
Version: 0.1.0
Summary: Open road data for autonomous driving on unstructured Indian roads.
Project-URL: Homepage, https://thirdeyelabs.ai
Project-URL: Dataset, https://huggingface.co/datasets/thirdeyelabs/indian-road-dataset
Author: Third Eye Labs
License: Apache-2.0
Keywords: adas,autonomous-driving,computer-vision,dataset,india,object-detection
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Science/Research
Classifier: License :: OSI Approved :: Apache Software License
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Topic :: Scientific/Engineering :: Image Recognition
Requires-Python: >=3.9
Provides-Extra: hf
Requires-Dist: datasets>=2.14.0; extra == 'hf'
Description-Content-Type: text/markdown

# thirdeyelabs

Open road data for autonomous driving on unstructured Indian roads.

Every major AV dataset (Waymo Open, nuScenes, Argoverse, PandaSet) was captured
on structured Western roads. This one was not.

```bash
pip install thirdeyelabs[hf]
```

```python
import thirdeyelabs as tel

tel.info()
# {'frames': 646014, 'clips': 8441, 'detections': 6896202, ...}

tel.INDIA_SPECIFIC
# ['autorickshaw', 'animal', 'vehicle_fallback']

ds = tel.load_dataset()
```

## The dataset

| | |
|---|---|
| Frames | 646,014 across 8,441 clips |
| Detections | 6,896,202 2D boxes |
| Segmentation | 1,290,463 masks |
| Classes | 12, BDD100K schema plus India-specific categories |
| Region | Delhi NCR, India |
| Sensor | Monocular dashcam, GPS per clip |
| License | CC BY 4.0 |

Three of the twelve categories (`autorickshaw`, `animal`, `vehicle_fallback`)
have no equivalent in COCO or BDD100K, so a detector trained on those
benchmarks cannot emit them at all.

Dataset: https://huggingface.co/datasets/thirdeyelabs/indian-road-dataset

## Status

Early. This release provides dataset access helpers and the taxonomy.
Simulation environments and benchmark suites are in development.

## License

Package: Apache-2.0. Dataset: CC BY 4.0.
