A multimodal question-answering dataset derived from the nuScenes-QA dataset for autonomous driving scenarios. It contains 2,229 training and 659 validation samples for day scenes, and 659 training and 659 validation samples for night scenes. The dataset was created by KevinNotSmile for evaluation in a research paper and was last updated on January 19, 2024.
Use Cases
- Training multimodal LLMs for visual question-answering based on autonomous driving scene data.
- Benchmarking model performance on day versus night scene understanding tasks.
- Evaluating modality adaptation methods in embodied AI systems based on the described multimodal QA tasks.
Strengths
- Provides a clear split of 2,229 training and validation samples for day scenes.
- Includes a distinct subset of 659 samples for night scenes, enabling analysis of lighting conditions.
- Created for a specific research purpose, indicating a focused design for multimodal QA evaluation.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown for the full dataset, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- Derived from the nuScenes-QA dataset.
- Collection Method
- Created for evaluation in the paper 'Modality Plug-and-Play: Elastic Modality Adaptation in Multimodal LLMs for Embodied AI'.
- Time Range
- null
- Freshness
- Last updated 2024-01-19 03:02:03; freshness should be verified.
- Geography
- null