Papers/2608.19380
🧪 Test?View on arXiv

CAViAR: A Causal Video Dataset for Fine-Grained Accident Reasoning in Real-World Scenarios

Not specified in the provided content

reasoningmultimodaldatasetautonomous driving
2608.19380
Builder Relevance
70%
2h ago

Abstract

CAViAR introduces a human-annotated dataset of real-world accident videos aimed at improving causal reasoning in autonomous driving systems.

Reality Card

Core Claim

CAViAR exposes a significant Perception--Reasoning Gap in current vision-language models, highlighting their inability to reliably map visible agent actions to responsibility categories in driving scenarios.

Method / Result

The dataset comprises 2,249 real-world accident videos with structured annotations.

Limitations

Current vision-language models show uneven performance, particularly in reasoning about accident types and responsibility.

Paper to code

Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.

No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.
← Back to all papers