metadata.json

{
  "@context": {
    "@vocab": "https://schema.org/",
    "@base": "https://mlcommons.org/croissant/",
    "ml": "http://mlcommons.org/schema#",
    "@language": "en"
  },
  "@type": "Dataset",
  "name": "Panel-Understanding-and-Operation",
  "description": "Panel Understanding and Operation (PUO) is a benchmark for evaluating vision-language models on panel understanding and operation tasks.",
  "url": "https://huggingface.co/datasets/Tele-AI-MAIL/Panel-Understanding-and-Operation",
  "version": "0.0.0",
  "keywords": ["vision-language models", "panel understanding", "instruction following", "privacy-preserving framework", "benchmark"],
  "license": "https://creativecommons.org/licenses/by/4.0/",
  "datePublished": "2025-05-14",
  "creator": {
    "@type": "Organization",
    "name": "Tele-AI-MAIL"
  },
  "includedInDataCatalog": {
    "@type": "DataCatalog",
    "name": "Hugging Face Datasets"
  },
  "distribution": [
    {
      "@type": "https://schema.org/FileObject",
      "name": "image.zip",
      "description": "Contains all panel images.",
      "contentUrl": "https://huggingface.co/datasets/Tele-AI-MAIL/Panel-Understanding-and-Operation/resolve/main/image.zip ",
      "encodingFormat": "application/zip",
      "sha256": "62436497a8ae518d8322cbdd1574be6fed3c8e12ed0019d6995c8d28d684e336"
    },
    {
      "@type": "https://schema.org/FileObject",
      "name": "label.zip",
      "description": "Corresponding label information, see Fig.5 in the paper.",
      "contentUrl": "https://huggingface.co/datasets/Tele-AI-MAIL/Panel-Understanding-and-Operation/resolve/main/label.zip ",
      "encodingFormat": "application/zip",
      "sha256": "25d8a9e2e31176dd045bf4c3fe8bac22385fcb64dde9138891918e22d0b23018"
    },
    {
      "@type": "https://schema.org/FileObject",
      "name": "instruction.zip",
      "description": "All QA pairs, see Fig.7 in the paper.",
      "contentUrl": "https://huggingface.co/datasets/Tele-AI-MAIL/Panel-Understanding-and-Operation/resolve/main/instruction.zip ",
      "encodingFormat": "application/zip",
      "sha256": "c94ebd08e4d06c4554da11ec328875888c800e2a64dcb431263c6f17358d6e1b"
    },
    {
      "@type": "https://schema.org/FileObject",
      "name": "split.json",
      "description": "Shows how training and test set are split.",
      "contentUrl": "https://huggingface.co/datasets/Tele-AI-MAIL/Panel-Understanding-and-Operation/resolve/main/split.json ",
      "encodingFormat": "application/json",
      "sha256": "13940f5231cd3a725e7d7253a14c3dc8dab32223671a8814518646ea8eac1360"
    }
  ],
  "isAccessibleForFree": true,
  "about": "A large-scale image-based dataset containing over 19k annotated panel images and 430k instruction-following QA pairs. Using this dataset. It can be used to fine-tune VLMs to perform key PUO tasks: panel description, element grounding, function estimation, and goal-based planning."
}