# Datasets
###### tags: `graduated` `dataset`
ref: https://blog.csdn.net/z704630835/article/details/99844183

## VQA v2
website: https://visualqa.org/
paper: https://arxiv.org/pdf/1612.00837.pdf

image_source: COCO
size: 1.1 Million (image, question) pairs


e.g.

## Visual-Genome (International Journal of Computer Vision 2017)

website: https://visualgenome.org/
paper: https://visualgenome.org/static/paper/Visual_Genome.pdf

e.g.

---

## VQA-E
> The `VQA-E` dataset is automatically derived from the popular VQA v2 dataset.

paper: https://arxiv.org/pdf/1803.07464.pdf
references: https://blog.csdn.net/z704630835/article/details/102721997
image_source: COCO

e.g.

---

## SNLI-VE: Visual Entailment Dataset
> `SNLI-VE` is built on top of `SNLI` and `Flickr30K`. The problem that VE is trying to solve is to reason about the relationship between an image premise $P_{image}$ and a text hypothesis $H_{text}$.

github: https://github.com/allenai/allennlp-models, https://github.com/necla-ml/SNLI-VE
image_source:

e.g.

## VQA-CP
> 基於 VQAv2 ，打亂training, testing dataset答案的分佈

paper: https://arxiv.org/pdf/1712.00377.pdf

## Visual7W
> 選擇題， 由 47,300 張 COCO，327,929 問答，1,311,756 個人工生成的多項選擇和 561,459 個對象基礎。

paper: http://ai.stanford.edu/~yukez/papers/cvpr2016.pdf

e.g.

## GQA (IEEE/CVF 2019)
> 新的VQA資料集，由 [COCO](https://cocodataset.org/#home), [Flickr](https://webscope.sandbox.yahoo.com/catalog.php?datatype=i&did=67) and [Visual
Genome](https://visualgenome.org/) 的圖片組成

paper: https://arxiv.org/pdf/1902.09506.pdf
website: https://cs.stanford.edu/people/dorarad/gqa/index.html

e.g.

## e-ViL(ICCV 2021)
> human-written NLEs (natural language explanations), provides a unified evaluation framework that is designed to be re-usable for future works

paper: https://openaccess.thecvf.com/content/ICCV2021/papers/Kayser_E-ViL_A_Dataset_and_Benchmark_for_Natural_Language_Explanations_in_ICCV_2021_paper.pdf


e.g.

## VCR (CVPR 2019)
website: https://visualcommonsense.com/
paper: https://arxiv.org/pdf/1811.10830.pdf

usage:

e.g.

## LVIS (CVPR 2019)
tag: `segmentation`
paper: https://arxiv.org/abs/1908.03195
e.g.

## MMDialog
github: https://github.com/victorsungo/MMDialog
paper:https://arxiv.org/abs/2211.05719

e.g.

## VQA-X
paper: https://arxiv.org/pdf/1711.07373.pdf
github:
drive:

## ACT-X (CVPR'14)
paper: https://ieeexplore.ieee.org/document/6909866

## TextVQA-X
github: https://github.com/amzn/explainable-text-vqa

## IconQA
website: https://iconqa.github.io/