README.md

metadata

dataset_info:
  features:
    - name: image
      dtype: image
    - name: segmentation
      dtype: image
    - name: captions
      sequence: string
  splits:
    - name: train
      num_bytes: 5999939927.78
      num_examples: 49999
  download_size: 5991450661
  dataset_size: 5999939927.78
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: cc-by-4.0
task_categories:
  - image-segmentation
  - image-to-text
  - text-to-image
language:
  - en
pretty_name: COCO 2017 segmentation dataset downsampled with captions
size_categories:
  - 10K<n<100K

📄 License and Attribution

This dataset is a downsampled version of the COCO 2017 dataset, tailored for segmentation tasks. It has the following fields:

• image: 256x256 image
• segmentation: 256x256 image. Each pixel encodes the class of that pixel. See class_names_dict.json for a legend.
• captions: a list of captions for the image, each by a different labeler.

Use the dataset as follows:

import requests
from datasets import load_dataset

ds = load_dataset("peteole/coco2017-segmentation", split="train")

# Optional: Load the class names as dict
url = "https://huggingface.co/datasets/peteole/coco2017-segmentation-50k-256x256/resolve/main/class_names_dict.json"
response = requests.get(url)
class_names_dict = response.json()

License

• License Type: Creative Commons Attribution 4.0 International (CC BY 4.0)
• License Details: This license permits redistribution, modification, and commercial use, provided that appropriate credit is given to the original creators.
• Original Dataset License: The original COCO 2017 dataset is licensed under CC BY 4.0.

Attribution

When using this dataset, please cite the original COCO dataset as follows:

Tsung-Yi Lin, Michael Maire, Serge Belongie, Lubomir Bourdev, Ross Girshick, James Hays, Pietro Perona, Deva Ramanan, C. Lawrence Zitnick, and Piotr Dollár. "Microsoft COCO: Common Objects in Context." In European Conference on Computer Vision, pp. 740–755. Springer, 2014.

For more information, visit the COCO dataset website.