metadata

language:
  - en
pretty_name: COCO2017
size_categories:
  - 100K<n<1M
task_categories:
  - image-to-text
task_ids:
  - image-captioning
tags:
  - coco
  - image-captioning
dataset_info:
  features:
    - name: license
      dtype: int64
    - name: file_name
      dtype: string
    - name: coco_url
      dtype: string
    - name: height
      dtype: int64
    - name: width
      dtype: int64
    - name: date_captured
      dtype: string
    - name: flickr_url
      dtype: string
    - name: image_id
      dtype: int64
    - name: ids
      sequence: int64
    - name: captions
      sequence: string
  splits:
    - name: train
      num_bytes: 64026361
      num_examples: 118287
    - name: validation
      num_bytes: 2684731
      num_examples: 5000
  download_size: 30170127
  dataset_size: 66711092

coco2017

Image-text pairs from MS COCO2017.

Data origin

Data originates from cocodataset.org
While coco-karpathy uses a dense format (with several sentences and sendids per row), coco-karpathy-long uses a long format with one sentence (aka caption) and sendid per row. coco-karpathy-long uses the first five sentences and therefore is five times as long as coco-karpathy.
- phiyodr/coco2017: One row corresponds one image with several sentences.
- phiyodr/coco2017-long: One row correspond one sentence (aka caption). There are 5 rows (sometimes more) with the same image details.

Format

DatasetDict({
    train: Dataset({
        features: ['license', 'file_name', 'coco_url', 'height', 'width', 'date_captured', 'flickr_url', 'image_id', 'ids', 'captions'],
        num_rows: 118287
    })
    validation: Dataset({
        features: ['license', 'file_name', 'coco_url', 'height', 'width', 'date_captured', 'flickr_url', 'image_id', 'ids', 'captions'],
        num_rows: 5000
    })
})

Usage

Download image data and unzip

cd PATH_TO_IMAGE_FOLDER

wget http://images.cocodataset.org/zips/train2017.zip
wget http://images.cocodataset.org/zips/val2017.zip
#wget http://images.cocodataset.org/annotations/annotations_trainval2017.zip # zip not needed: everything you need is in load_dataset("phiyodr/coco2017")

unzip train2017.zip
unzip val2017.zip

Load dataset in Python

import os
from datasets import load_dataset
PATH_TO_IMAGE_FOLDER = "COCO2017"

def create_full_path(example):
    """Create full path to image using `base_path` to COCO2017 folder."""
    example["image_path"] = os.path.join(PATH_TO_IMAGE_FOLDER, example["file_name"])
    return example

dataset = load_dataset("phiyodr/coco2017")
dataset = dataset.map(create_full_path)