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.