Spaces:
Running on Zero
Running on Zero
| 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. | |