---
dataset_info:
  features:
  - name: id
    dtype: string
  - name: image
    dtype: image
  - name: obj_text
    sequence: string
  - name: ref_ids
    sequence: int64
  - name: json_data
    list:
    - name: ref_id
      dtype: int64
    - name: text
      sequence: string
  - name: depth_caption
    dtype: string
  - name: pred_res
    sequence: int64
  splits:
  - name: train
    num_bytes: 12965187.0
    num_examples: 100
  download_size: 12935674
  dataset_size: 12965187.0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---


# Som_bench_refcocog_refseg Dataset

This dataset is a processed version of the RefCOCOg dataset and is intended to be used as part of a benchmark, specifically mirroring the data splits and format used in the [Set-of-Mark (SoM)](https://github.com/microsoft/SoM/tree/main/benchmark) benchmark.  It is designed for evaluating visual grounding and related tasks.

**Original Dataset:**

This dataset is based on the [RefCOCOg](https://github.com/lichengunc/refer) dataset.  Please refer to the original RefCOCOg dataset for its terms of use and licensing.

**Benchmark Reference:**

This dataset follows the benchmark setup described in the following repository:

*   [Set-of-Mark (SoM) Benchmark](https://github.com/microsoft/SoM/tree/main/benchmark)

**Citation (SoM):**

If you use this *benchmark setup* in your research, please cite the following paper:

```bibtex
@article{{yang2023setofmark,
    title={{Set-of-Mark Prompting Unleashes Extraordinary Visual Grounding in GPT-4V}},
    author={{Jianwei Yang and Hao Zhang and Feng Li and Xueyan Zou and Chunyuan Li and Jianfeng Gao}},
    journal={{arXiv preprint arXiv:2310.11441}},
    year={{2023}},
}}