adaspun

ADASPUN

ADASPUN is a research-oriented Python package for ADAS perception, stereo depth, risk estimation, QUBO warning selection, and NeuroSentinel-4D++ V9 execution.

Install

Install from PyPI:

pip install adaspun

For local development with YOLO extras:

python -m pip install -e ".[yolo]"

Quick checks

adaspun-doctor
adaspun-v9-robust --help
adaspun-native-video --help

Robust V9 source-of-truth runner

The robust V9 command delegates to a local NeuroSentinel V9 research workspace. A full V9 run requires local assets such as the V9 script, helper backbone, model weights, and input videos.

Example:

adaspun-v9-robust --project-root D:\Puneeth_Adas --v9-script D:\Puneeth_Adas\final_video_pipeline\run_neurosentinel_ddpm_ensemble_v9_sota_clean_ui_DISTANCE_LABELS.py --cuda --doctor-first --copy-output D:\Puneeth_Adas\outputs\adaspun_v9_demo.mp4

Native modular commands

adaspun-native-video --help
adaspun-risk-smoke --help
adaspun-qubo-smoke --help

Research preview note

ADASPUN is currently a research preview package. The modular native commands are included for package development and experimentation. The robust V9 command is a bridge to the complete local NeuroSentinel V9 source-of-truth pipeline and does not bundle model weights, datasets, videos, or private research assets.

Citation

If you use ADASPUN in academic work, cite the repository/package metadata in CITATION.cff.

Python API

from adaspun import ADASPUNRunner

runner = ADASPUNRunner(model=r"D:\path\to\best.pt", cuda=True)
result = runner.run_video("road.mp4", "output.mp4", max_frames=120)
print(result.output_video)

See docs/python_api_usage.md for more examples.

Hosted model weights

ADASPUN does not bundle trained .pt weights in the PyPI wheel. Use local .pt files or host weights on Hugging Face/GitHub Releases and update the model registry. See docs/hosted_models.md.