Skip to content

Dataset · nvidia

NVIDIA traffic anomaly video and QA dataset

nvidia/PhysicalAI-Traffic-Anomaly-Reasoning

Dataset with videos and text questions from NVIDIA for road anomaly analysis, accepted at NeurIPS 2026.

What's inside

The collection contains 10k to 100k examples featuring English video and text data. The data is designed for visual question answering (VQA) and video-to-text generation, complete with loader configurations and Croissant metadata. This work was accepted into the NeurIPS 2026 Datasets and Benchmarks track and presented at the AI City Challenge workshop at ECCV 2026.

What you can do with it

  • Training vision-language models (VLM) to recognize road emergencies
  • Evaluating video analytics systems in traffic VQA tasks
Tasks
Video text → text, Visual question answering
Languages
English
Size
10K–100K rows
Downloads this month
1.8K
Likes
23

How to load it

from datasets import load_dataset

ds = load_dataset("nvidia/PhysicalAI-Traffic-Anomaly-Reasoning")