DOCSEARCHbyAlgolia
Submit Docs

TensorFlow Datasets (TFDS)

repository·master·Indexed 26 days ago

https://github.com/tensorflow/datasets

A collection of ready-to-use public datasets provided as tf.data.Dataset objects for high performance and ease of use within the TensorFlow ecosystem. It includes a wide variety of datasets for tasks such as text classification, question-answering, and facial landmark detection, and provides the tfds.load() API for constructing input pipelines.

Tokens
87.3K
Snippets
132
Records
686
Agent score
88%

What's inside tensorflow-datasets

  1. Overview of the Schema-Guided Dialogue (SGD) dataset

    master

    The Schema-Guided Dialogue (SGD) dataset contains over 20,000 annotated multi-domain, task-oriented conversations between humans and virtual assistants. It covers 20 domains (e.g., banks, events, media, calendar, travel, weather) and includes multiple APIs per domain with varying interfaces.

    Key use cases include:

    • Intent prediction
    • Slot filling
    • Dialogue state tracking
    • Policy imitation learning
    • Language generation
    • User simulation learning

    The evaluation set includes unseen domains and services to facilitate zero-shot and few-shot performance quantification.

  4. Overview of the QA4MRE dataset

    master

    The QA4MRE dataset is designed for research in question answering and reading comprehension. It consists of supporting passages paired with a set of questions. For each question, multiple answer options are provided, with only one being correct.

    Key components include:

    • Main Track: Contains training and test datasets.
    • Pilot Studies: Includes gold standard documents for Alzheimer's data and entrance exams data.

  5. Overview of the PlantVillage dataset

    master

    The PlantVillage dataset contains 54,303 images of healthy and unhealthy leaves, categorized into 38 different species and diseases.

    Note on Data Provenance:

    • The dataset is sourced from a republished unaugmented version via a research paper, as the original source (plantvillage.org) is unavailable.
    • Images with the Background_without_leaves label have been removed to maintain consistency with the original dataset's scope.

    References:

  6. Overview of the Wake Vision dataset

    master
    Wake Vision is a large-scale, high-quality dataset designed for tinyML applications, containing over 6 million images. It provides annotations indicating the presence of a person in each image. The dataset includes a fine-grained benchmark to evaluate model fairness and robustness across several dimensions: perceived gender, perceived age, subject distance, lighting conditions, and depictions.

  8. Overview of the SAMSum dataset

    master

    The SAMSum dataset is a corpus containing over 16,000 chat dialogues, each paired with a manually annotated summary. It is designed for tasks involving dialogue summarization.

    Each example in the dataset contains the following features:

    • dialogue: The text content of the chat dialogue.
    • summary: A human-written summary of the dialogue.
    • id: A unique identifier for the example.

  9. Overview of ImageNet-Sketch dataset

    master
    ImageNet-Sketch is a dataset containing 50,889 black and white sketch images. It covers 1,000 ImageNet classes, with 50 images provided for each class. The images were originally sourced from Google Image Search using the query format "sketch of __" and were manually filtered. For classes that lacked 50 high-quality images, additional samples were generated via flipping or rotation.