Spaces:
Running on Zero
Running on Zero
File size: 2,083 Bytes
f223812 eaaedd2 f223812 eaaedd2 f223812 eaaedd2 328c703 f223812 eaaedd2 0d65642 eaaedd2 0d65642 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 | ---
title: Cosmos Sentinel
emoji: 🚦
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 6.8.0
python_version: '3.12'
app_file: app.py
pinned: false
license: apache-2.0
short_description: BADAS + Cosmos Reason 2 ZeroGPU demo.
models:
- nexar-ai/BADAS-Open
- nvidia/Cosmos-Reason2-8B
- nvidia/Cosmos-Predict2-2B
tags:
- autonomous-driving
- collision-prediction
- traffic-safety
- video-understanding
- zero-gpu
---
# Cosmos Sentinel
Cosmos Sentinel is a traffic-safety demo built around:
- **BADAS** for predictive collision gating
- **Cosmos Reason 2** for causal incident narration and severity labeling
- **Cosmos Predict 2.5** for optional future continuation rollouts in the full project repo
This Hugging Face Space is implemented as a **Gradio app** so it is compatible with **ZeroGPU**.
## What this Space includes
- a polished overview of the project
- a cached sample video for immediate testing
- BADAS + Cosmos Reason 2 live pipeline execution in Gradio
- manual Cosmos Predict continuation rollouts
- model warmup hooks for ZeroGPU-backed startup
- supplementary links to the main papers, repos, and official docs
## Full project repository
- GitHub: https://github.com/Ryukijano/Nvidia-Cosmos-Cookoff
## Supplementary materials
- NVIDIA Cosmos overview: https://docs.nvidia.com/cosmos/latest/introduction.html
- Cosmos Reason 2 docs: https://docs.nvidia.com/cosmos/latest/reason2/index.html
- Cosmos Reason 2 repo: https://github.com/nvidia-cosmos/cosmos-reason2
- Cosmos Predict 2.5 repo: https://github.com/nvidia-cosmos/cosmos-predict2.5
- Cosmos Cookbook: https://nvidia-cosmos.github.io/cosmos-cookbook/index.html
- BADAS paper: https://arxiv.org/abs/2510.14876
- Nexar dataset paper: https://openaccess.thecvf.com/content/CVPR2025W/WAD/papers/Moura_Nexar_Dashcam_Collision_Prediction_Dataset_and_Challenge_CVPRW_2025_paper.pdf
## Note
The full local Streamlit dashboard remains in the GitHub repo, but this Space now mirrors the core pipeline in a Gradio-first interface tailored for Hugging Face Spaces and ZeroGPU deployment.
|