Explore @google-research-datasets Open Source projects

Google Research Datasets (@google-research-datasets)

google-research-datasets

Stars
12,798
Global Org. Rank 1,806 (Top 0.6 %)
Registered almost 9 years ago
Most used languages

Python
67.6 %

Jupyter Notebook
18.9 %

C++
2.7 %

Starlark 2.7 %

Shell
2.7 %

Macaulay2 2.7 %

Makefile 2.7 %
Location 🇺🇸 United States
Country Total Rank 850
Country Ranking

Macaulay2 2

Jupyter Notebook
103

Starlark 129

Makefile 191

Python
386

Shell
468

C++
2,128

Objectron

Objectron is a dataset of short, object-centric video clips. In addition, the videos also contain AR session metadata including camera poses, sparse point-clouds and planes. In each video, the camera moves around and above the object and captures it from different views. Each object is annotated with a 3D bounding box. The 3D bounding box describes the object’s position, orientation, and dimensions. The dataset contains about 15K annotated video clips and 4M annotated images in the following categories: bikes, books, bottles, cameras, cereal boxes, chairs, cups, laptops, and shoes

Jupyter Notebook

wit

WIT (Wikipedia-based Image Text) Dataset is a large multimodal multilingual dataset comprising 37M+ image-text sets with 11M+ unique images across 100+ languages.

natural-questions

Natural Questions (NQ) contains real user questions issued to Google search, and answers found from Wikipedia by annotators. NQ is designed for the training and evaluation of automatic question answering systems.

paws

This dataset contains 108,463 human-labeled and 656k noisily labeled pairs that feature the importance of modeling structure, context, and word order information for the problem of paraphrase identification.

dstc8-schema-guided-dialogue

The Schema-Guided Dialogue Dataset

conceptual-captions

Conceptual Captions is a dataset containing (image-URL, caption) pairs designed for the training and evaluation of machine learned image captioning systems.

ToTTo

ToTTo is an open-domain English table-to-text dataset with over 120,000 training examples that proposes a controlled generation task: given a Wikipedia table and a set of highlighted table cells, produce a one-sentence description. We hope it can serve as a useful research benchmark for high-precision conditional text generation.

conceptual-12m

Conceptual 12M is a dataset containing (image-URL, caption) pairs collected for vision-and-language pre-training.

tydiqa

TyDi QA contains 200k human-annotated question-answer pairs in 11 Typologically Diverse languages, written without seeing the answer and without the use of translation, and is designed for the training and evaluation of automatic question answering systems. This repository provides evaluation code and a baseline system for the dataset.

wiki-reading

This repository contains the three WikiReading datasets as used and described in WikiReading: A Novel Large-scale Language Understanding Task over Wikipedia, Hewlett, et al, ACL 2016 (the English WikiReading dataset) and Byte-level Machine Reading across Morphologically Varied Languages, Kenter et al, AAAI-18 (the Turkish and Russian datasets).

hiertext

The HierText dataset contains ~12k images from the Open Images dataset v6 with large amount of text entities. We provide word, line and paragraph level annotations.

Jupyter Notebook

coarse-discourse

A large corpus of discourse annotations and relations on ~10K forum threads.

simulated-dialogue

gap-coreference

GAP is a gender-balanced dataset containing 8,908 coreference-labeled pairs of (ambiguous pronoun, antecedent name), sampled from Wikipedia for the evaluation of coreference resolution in practical applications.

KELM-corpus

Taskmaster

Please see the readme file as well as our 2019 EMNLP paper linked here -->

dakshina

The Dakshina dataset is a collection of text in both Latin and native scripts for 12 South Asian languages. For each language, the dataset includes a large collection of native script Wikipedia text, a romanization lexicon of words in the native script with attested romanizations, and some full sentence parallel data in both a native script of the language and the basic Latin alphabet.

word_sense_disambigation_corpora

SemCor and Masc documents annotated with NOAD word senses.

cvss

CVSS: A Massively Multilingual Speech-to-Speech Translation Corpus

C4_200M-synthetic-dataset-for-grammatical-error-correction

This dataset contains synthetic training data for grammatical error correction. The corpus is generated by corrupting clean sentences from C4 using a tagged corruption model. The approach and the dataset are described in more detail by Stahlberg and Kumar (2021) (https://www.aclweb.org/anthology/2021.bea-1.4/)

Nutrition5k

Detailed visual + nutritional data for over 5,000 plates of food.

boolean-questions

MAVE

The dataset contains 3 million attribute-value annotations across 1257 unique categories on 2.2 million cleaned Amazon product profiles. It is a large, multi-sourced, diverse dataset for product attribute extraction study.

wiki-split

One million English sentences, each split into two sentences that together preserve the original meaning, extracted from Wikipedia edits.

sentence-compression

Large corpus of uncompressed and compressed sentences from news articles.

tpu_graphs

QED

QED: A Framework and Dataset for Explanations in Question Answering

presto

A Multilingual Dataset for Parsing Realistic Task-Oriented Dialogs

RxR

Room-across-Room (RxR) is a large-scale, multilingual dataset for Vision-and-Language Navigation (VLN) in Matterport3D environments. It contains 126k navigation instructions in English, Hindi and Telugu, and 126k navigation following demonstrations. Both annotation types include dense spatiotemporal alignments between the text and the visual perceptions of the annotators

wiki-atomic-edits

A dataset of atomic wikipedia edits containing insertions and deletions of a contiguous chunk of text in a sentence. This dataset contains ~43 million edits across 8 languages.

clang8

cLang-8 is a dataset for grammatical error correction.

seahorse

Seahorse is a dataset for multilingual, multi-faceted summarization evaluation. It consists of 96K summaries with human ratings along 6 quality dimensions: comprehensibility, repetition, grammar, attribution, main idea(s), and conciseness, covering 6 languages, 9 systems and 4 datasets.

query-wellformedness

25,100 queries from the Paralex corpus (Fader et al., 2013) annotated with human ratings of whether they are well-formed natural language questions.

xsum_hallucination_annotations

Faithfulness and factuality annotations of XSum summaries from our paper "On Faithfulness and Factuality in Abstractive Summarization" (https://www.aclweb.org/anthology/2020.acl-main.173.pdf).

videoCC-data

VideoCC is a dataset containing (video-URL, caption) pairs for training video-text machine learning models. It is created using an automatic pipeline starting from the Conceptual Captions Image-Captioning Dataset.

TimeDial

Temporal Commonsense Reasoning in Dialog

vrdu

We identify the desiderata for a comprehensive benchmark and propose Visually Rich Document Understanding (VRDU). VRDU contains two datasets that represent several challenges: rich schema including diverse data types, complex templates, and diversity of layouts within a single document type.

uninum

A database of number names for 186 languages, locales, and scripts

screen_qa

ScreenQA dataset was introduced in the "ScreenQA: Large-Scale Question-Answer Pairs over Mobile App Screenshots" paper. It contains ~86K question-answer pairs collected by human annotators for ~35K screenshots from Rico. It should be used to train and evaluate models capable of screen content understanding via question answering.

TextNormalizationCoveringGrammars

Covering grammars for English and Russian text normalization

Disfl-QA

A Benchmark Dataset for Understanding Disfluencies in Question Answering

relation-extraction-corpus

Automatically exported from code.google.com/p/relation-extraction-corpus

WikipediaHomographData

Labeled data for homograph disambiguation

scin

The SCIN dataset contains 10,000+ images of dermatology conditions, crowdsourced with informed consent from US internet users. Contributions include self-reported demographic and symptom information and dermatologist labels. The dataset also contains estimated Fitzpatrick skin type and Monk Skin Tone.

Jupyter Notebook

GSM-IC

Grade-School Math with Irrelevant Context (GSM-IC) benchmark is an arithmetic reasoning dataset built upon GSM8K, by adding irrelevant sentences in problem descriptions. GSM-IC is constructed to evaluate the distractibility of language models.

Crisscrossed-Captions

Extended Intramodal and Intermodal Semantic Similarity Judgments for MS-COCO

bam

synthetic-fur

A procedurally generated synthetic fur dataset with conditional inputs for machine learning and neural rendering.

wiki-links

Automatically exported from code.google.com/p/wiki-links

Synthetic-Persona-Chat

The Synthetic-Persona-Chat dataset is a synthetically generated persona-based dialogue dataset. It extends the original Persona-Chat dataset.

swim-ir

SWIM-IR is a Synthetic Wikipedia-based Multilingual Information Retrieval training set with 28 million query-passage pairs spanning 33 languages, generated using PaLM 2 and summarize-then-ask prompting.

Attributed-QA

We believe the ability of an LLM to attribute the text that it generates is likely to be crucial for both system developers and users in information-seeking scenarios. This release consists of human-rated system outputs for a new question-answering task, Attributed Question Answering (AQA).

uibert

It includes two datasets that are used in the downstream tasks for evaluating UIBert: App Similar Element Retrieval data and Visual Item Selection (VIS) data. Both datasets are written TFRecords.

sanpo_dataset

screen2words

The dataset includes screen summaries that describes Android app screenshot's functionalities. It is used for training and evaluation of the screen2words models (our paper accepted by UIST'21 will be linked soon).

clay

The dataset includes UI object type labels (e.g., BUTTON, IMAGE, CHECKBOX) that describes the semantic type of an UI object on Android app screenshots. It is used for training and evaluation of the screen layout denoising models (paper will be linked soon).

noun-verb

This dataset contains naturally-occurring English sentences that feature non-trivial noun-verb ambiguity.

NewSHead

The NewSHead dataset is a multi-doc headline dataset used in NHNet for training a headline summarization model.

eev

The Evoked Expressions in Video dataset contains videos paired with the expected facial expressions over time exhibited by people reacting to the video content.

TF-IDF-IIF-top100-wordlists

These are lists for a variety of languages containing words that are distinctive to each language.

QAmeleon

QAmeleon introduces synthetic multilingual QA data using PaLM, a 540B large language model. This dataset was generated by prompt tuning PaLM with only five examples per language. We use the synthetic data to finetune downstream QA models leading to improved accuracy in comparison to English-only and translation-based baselines.

Hinglish-TOP-Dataset

Consists of the largest (10K) human annotated code-switched semantic parsing dataset & 170K generated utterance using the CST5 augmentation technique. Queries are derived from TOPv2, a multi-domain task oriented semantic parsing dataset. Tests suggest that with CST5, up to 20x less labeled data can achieve the same semantic parsing performance.

discofuse

NewsQuizQA

NewsQuizQA is a quiz-style question-answer dataset used for generating quiz questions about the news

turkish-treebanks

A human-annotated morphosyntactic treebank for Turkish.

seegull

SeeGULL is a broad-coverage stereotype dataset in English containing stereotypes about identity groups spanning 178 countries across 8 different geo-political regions across 6 continents, as well as state-level identities within the US and India.

global_streamflow_model_paper

Jupyter Notebook

Image-Caption-Quality-Dataset

A dataset of crowdsourced ratings for machine-generated image captions

eth_py150_open

A redistributable subset of the ETH Py150 corpus [https://www.sri.inf.ethz.ch/py150], introduced in the ICML 2020 paper 'Learning and Evaluating Contextual Embedding of Source Code' [https://proceedings.icml.cc/static/paper_files/icml/2020/5401-Paper.pdf].

MultiReQA

We are creating a challenging new benchmark MultiReQA: A Cross-Domain Evaluation for Retrieval Question Answering Models. Retrieval question answering (ReQA) is the task of retrieving a sentence-level answer to a question from an open corpus. MultiReQA is a new multi-domain ReQA evaluation suite composed of eight retrieval QA tasks drawn from publicly available QA datasets from the MRQA shared task. We believe that MultiReQA tests retrieval QA models’ ability to perform domain transfer tasks. This repository hosts the codes to convert existing QA datasets from MRQA shared task to the format of MultiReQA benchmark, as well as the sentence boundary annotations for QA datasets to exactly reproduce our work. Note that we are not redistributing the content in the original datasets available on MRQA share task, but just the sentence boundary annotations.

seq2act

This repository contains the opensource version of the datasets were used for different parts of training and testing of models that ground natural language to UI actions as described in the paper: "Mapping Natural Language Instructions to Mobile UI Action Sequences" by Yang Li, Jiacong He, Xin Zhou, Yuan Zhang, and Jason Baldridge, which is accepted in 2020 Annual Conference of the Association for Computational Linguistics (ACL 2020)

AIS

AIS is an evaluation framework for assessing whether the output of natural language models only contains information about the external world that is verifiable in source documents, or "Attributable to Identified Sources".

ccpe

A dataset consisting of 502 English dialogs with 12,000 annotated utterances between a user and an assistant discussing movie preferences in natural language. It was collected using a Wizard-of-Oz methodology between two paid crowd-workers, where one worker plays the role of an 'assistant', while the other plays the role of a 'user'. The 'assistant' elicits the 'user’s' preferences about movies following a Coached Conversational Preference Elicitation (CCPE) method. The assistant asks questions designed to minimize the bias in the terminology the 'user' employs to convey his or her preferences as much as possible, and to obtain these preferences in natural language. Each dialog is annotated with entity mentions, preferences expressed about entities, descriptions of entities provided, and other statements of entities.

wikifact

Wikipedia based dataset to train relationship classifiers and fact extraction models

great

The dataset for the variable-misuse task, used in the ICLR 2020 paper 'Global Relational Models of Source Code' [https://openreview.net/forum?id=B1lnbRNtwr]

nyt-salience

Automatically exported from code.google.com/p/nyt-salience

dices-dataset

This repository contains two datasets with multi-turn adversarial conversations generated by human agents interacting with a dialog model and rated for safety by two corresponding diverse rater pools.

indic-gen-bench

IndicGenBench is a high-quality, multilingual, multi-way parallel benchmark for evaluating Large Language Models (LLMs) on 4 user-facing generation tasks across a diverse set 29 of Indic languages covering 13 scripts and 4 language families.

WebRED

WebRED is a large and diverse manually annotated dataset for extracting relationships from a variety of text found on the World Wide Web.

Video-Timeline-Tags-ViTT

A collection of videos annotated with timelines where each video is divided into segments, and each segment is labelled with a short free-text description

answer-equivalence-dataset

This dataset contains human judgements about answer equivalence. The data is based on SQuAD (Stanford Question Answering Dataset), and contains 9k human judgements of answer candidates generated by Albert on the SQuAD train set, and an additional 14k human judgements for answer candidates produced by BiDAF, Luke, and XLNet on the SQuAD dev set.

Jupyter Notebook

circa

Circa (meaning ‘approximately’) dataset aims to help machine learning systems to solve the problem of interpreting indirect answers to polar questions. The dataset contains pairs of yes/no questions and indirect answers, together with annotations for the interpretation of the answer. The data is collected in 10 different social conversation situations (eg. food preferences of a friend).

rico_semantics

Consists of ~500k human annotations on the RICO dataset identifying various icons based on their shapes and semantics, and associations between selected general UI elements and their text labels. Annotations also include human annotated bounding boxes which are more accurate and have a greater coverage of UI elements.

screen_annotation

The Screen Annotation dataset consists of pairs of mobile screenshots and their annotations. The annotations are in text format, and describe the UI elements present on the screen: their type, location, OCR text and a short description. It has been introduced in the paper `ScreenAI: A Vision-Language Model for UI and Infographics Understanding`.

distribution-over-quantities

birds-to-words

widget-caption

The dataset includes widget captions that describes UI element's functionalities. It is used for training and evaluation of the widget captioning model (please see the EMNLP'20 paper: https://arxiv.org/abs/2010.04295).

common-crawl-domain-names

Corpus of domain names scraped from Common Crawl and manually annotated to add word boundaries (e.g. "commoncrawl" to "common crawl").

2.5vrd

This dataset contains about 110k images annotated with the depth and occlusion relationships between arbitrary objects. It enables research on the 2.5D Visual Relationship Detection (2.5VRD) introduced in https://arxiv.org/abs/2104.12727.

DaTaSeg-Objects365-Instance-Segmentation

We release the DaTaSeg Objects365 Instance Segmentation Dataset introduced in the DaTaSeg paper, which can be used as an evaluation benchmark for weakly or semi supervised segmentation.

Jupyter Notebook

maverics

MAVERICS (Manually-vAlidated Vq^2a Examples fRom Image-Caption datasetS) is a suite of test-only benchmarks for visual question answering (VQA).

PropSegmEnt

PropSegmEnt is an annotated dataset for segmenting English text into propositions, and recognizing proposition-level entailment relations - whether a different, related document entails each proposition, contradicts it, or neither. It consists of clusters of closely related documents from the news and Wikipedia domains.

lareqa

LAReQA is a challenging benchmark for evaluating language agnostic answer retrieval from a multilingual candidate pool. This repository contains a dataset we release as part of the LAReQA evaluation.

recognizing-multimodal-entailment

The dataset consists of public social media url pairs and the corresponding entailment label for an external conference (ACL 2021). Each url contains a post with both linguistic (text) and visual (image) content. Entailment labels are human annotated through Google Crowdsource.

Jupyter Notebook

Textual-Entailment-New-Protocols

This data release is meant to accompany and document the paper: https://arxiv.org/abs/2004.11997 Collecting Entailment Data for Pretraining: New Protocols and Negative Results by Samuel R. Bowman, Jennimaria Palomaki, Livio Baldini Soares, and Emily Pitler

nlp-fairness-for-india

Contains data resources to replicate results from the paper “Re-contextualizing Fairness in NLP: The Case of India”.

adversarial-nibbler

This dataset contains results from all rounds of Adversarial Nibbler. This data includes adversarial prompts fed into public generative text2image models and validations for unsafe images. There will be two sets of data: all prompts submitted and all prompts attempted (sent to t2i models but not submitted as unsafe).

aquamuse

AQuaMuSe is a novel scalable approach to automatically mine dual query based multi-document summarization datasets for extractive and abstractive summaries using question answering dataset (Google Natural Questions) and large document corpora (Common Crawl)

maxm

MaXM is a suite of test-only benchmarks for multilingual visual question answering in 7 languages: English (en), French (fr), Hindi (hi), Hebrew (iw), Romanian (ro), Thai (th), and Chinese (zh).

sense-anaphora

Publicly released data: sense anaphora annotations.