Datasets:
File size: 2,812 Bytes
ba8e465 149f228 8456bbb ba8e465 149f228 4b6635d 149f228 8456bbb | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 | ---
dataset_info:
features:
- name: images
list:
image:
decode: false
- name: image_group_id
dtype: string
- name: qa
list:
- name: view
dtype: string
- name: inputs
dtype: string
- name: choices_list
list: string
- name: labels
dtype: int64
- name: metadata
dtype: string
splits:
- name: test
num_bytes: 710395695
num_examples: 3980
- name: train
num_bytes: 1577067854
num_examples: 8526
- name: validation
num_bytes: 467812623
num_examples: 2566
download_size: 2755538492
dataset_size: 2755276172
configs:
- config_name: default
data_files:
- split: test
path: data/test-*
- split: train
path: data/train-*
- split: validation
path: data/validation-*
license: other
source_datasets:
- jxu124/refcoco
- jxu124/refcocoplus
- jxu124/refcocog
- Untitled111/refcoco-train2014-assets
tags:
- decision-model
- grounding
language:
- en
task_categories:
- multiple-choice
- visual-question-answering
license_name: mixed-flickr-and-original-source-terms
license_link: https://cocodataset.org/#termsofuse
---
# refcoco-grounding
Human referring expressions from RefCOCO, RefCOCO+ and RefCOCOg, joined to COCO2014.
One shared image cache and mirror; views retain their native question identities.
The union of all held-out image IDs is reserved from training across variants;
overlapping validation/test images stay in test only. Original split names and
per-image Flickr licenses are recorded without excluding noncommercial or
unresolved sources. Original images are redistributed unchanged; optional pixel
augmentations remain subject to each image's original terms.
Only native COCO object boxes supply options; no invented distractors.
Original data: [jxu124/refcoco](https://huggingface.co/datasets/jxu124/refcoco), [jxu124/refcocoplus](https://huggingface.co/datasets/jxu124/refcocoplus), [jxu124/refcocog](https://huggingface.co/datasets/jxu124/refcocog), [Untitled111/refcoco-train2014-assets](https://huggingface.co/datasets/Untitled111/refcoco-train2014-assets). Repackaged as parquet for [tasksource](https://github.com/sileod/tasksource) by [scripts/repackage_dataset/grounding.py](https://github.com/sileod/tasksource/blob/main/scripts/repackage_dataset/grounding.py#L88).
Images are stored once with grouped `qa` records: train: 100,000 questions / 8,526 image groups, validation: 24,527 questions / 2,566 image groups, test: 30,903 questions / 3,980 image groups.
[Conversion provenance](provenance.json): source revisions, hashes, configuration and exclusions.
[Original referring-expression release](https://github.com/lichengunc/refer) · [Coarse recast audit](https://github.com/sileod/tasksource/blob/main/docs/jev/audits/refcoco-grounding-recast.json).
|