Ego3D-Bench: Vision-Language Model Spatial Reasoning in Ego-Centric Multi-View Scenes

Ego3D-Bench is a benchmark for evaluating spatial understanding in ego-centric multi-view scenarios. Images are sourced from three autonomous driving datasets: NuScenes, Argoverse, and Waymo. The benchmark was created by author vbdai and last updated on 2026-01-26.

Use Cases

• Benchmarking cross-view reasoning capabilities of vision-language models based on multi-view image data.
• Evaluating spatial understanding from an ego-centric perspective based on the defined question types.
• Training models for scene comprehension in autonomous driving contexts based on images from established driving datasets.

Strengths

• Images are sourced from three established autonomous driving datasets: NuScenes, Argoverse, and Waymo.
• Questions are specifically designed to require cross-view reasoning.
• Last updated on 2026-01-26, indicating recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Images collected from NuScenes, Argoverse, and Waymo datasets.

Collection Method

Benchmark designed and questions created for evaluating spatial understanding.

Time Range

null

Freshness

Last updated 2026-01-26 23:16:43.

Geography

null