---
pretty_name: Point-It-Out (subset mirror)
task_categories:
- visual-question-answering
language:
- en
---

# Point-It-Out — subset mirror

A partial mirror of the data released with **Point-It-Out: Benchmarking Embodied Reasoning for
Vision Language Models in Multi-Stage Visual Grounding** ([arXiv:2509.25794](https://arxiv.org/abs/2509.25794)).

- Original source: <https://github.com/xavihart/PIO> (`data/`)
- `s1s2.json`: identical to the upstream `data/s1s2.json`.
- `images_s1s2/`: 132 of the upstream `data/images_s1s2/` images, byte-identical to the originals. Only the
  images referenced by our evaluation subsets (affordance, attribute reference, ordinal reference, safety) are included.

The upstream repository does not declare a license. All rights remain with the original authors; this mirror is
provided for evaluation convenience only. For the full dataset, use the upstream repository.

If you use this data, please cite the original paper.