Dataset Card for Visual Genome

Table of Contents

• Table of Contents
• Dataset Description
  • Dataset Summary
  • Dataset Preprocessing
  • Supported Tasks and Leaderboards
  • Languages
• Dataset Structure
  • Data Instances
  • Data Fields
  • Data Splits
• Dataset Creation
  • Curation Rationale
  • Source Data
  • Annotations
  • Personal and Sensitive Information
• Considerations for Using the Data
  • Social Impact of Dataset
  • Discussion of Biases
  • Other Known Limitations
• Additional Information
  • Dataset Curators
  • Licensing Information
  • Citation Information
  • Contributions

Dataset Description

• Homepage: https://homes.cs.washington.edu/~ranjay/visualgenome/
• Repository:
• Paper: https://doi.org/10.1007/s11263-016-0981-7
• Leaderboard:
• Point of Contact: ranjaykrishna [at] gmail [dot] com

Dataset Summary

Visual Genome is a dataset, a knowledge base, an ongoing effort to connect structured image concepts to language.

From the paper:

Despite progress in perceptual tasks such as image classification, computers still perform poorly on cognitive tasks such as image description and question answering. Cognition is core to tasks that involve not just recognizing, but reasoning about our visual world. However, models used to tackle the rich content in images for cognitive tasks are still being trained using the same datasets designed for perceptual tasks. To achieve success at cognitive tasks, models need to understand the interactions and relationships between objects in an image. When asked “What vehicle is the person riding?”, computers will need to identify the objects in an image as well as the relationships riding(man, carriage) and pulling(horse, carriage) to answer correctly that “the person is riding a horse-drawn carriage.”

Visual Genome has:

• 108,077 image
• 5.4 Million Region Descriptions
• 1.7 Million Visual Question Answers
• 3.8 Million Object Instances
• 2.8 Million Attributes
• 2.3 Million Relationships

From the paper:

Our dataset contains over 108K images where each image has an average of 35 objects, 26 attributes, and 21 pairwise relationships between objects. We canonicalize the objects, attributes, relationships, and noun phrases in region descriptions and questions answer pairs to WordNet synsets.

Dataset Preprocessing

Supported Tasks and Leaderboards

Languages

All of annotations use English as primary language.

Dataset Structure

Data Instances

When loading a specific configuration, users has to append a version dependent suffix:

from datasets import load_dataset
load_dataset("visual_genome", "region_description_v1.2.0")

region_descriptions

An example of looks as follows.

{
  "image": <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=800x600 at 0x7F2F60698610>,
  "image_id": 1,
  "url": "https://cs.stanford.edu/people/rak248/VG_100K_2/1.jpg",
  "width": 800,
  "height": 600,
  "coco_id": null,
  "flickr_id": null,
  "regions": [
    {
      "region_id": 1382,
      "image_id": 1,
      "phrase": "the clock is green in colour",
      "x": 421,
      "y": 57,
      "width": 82,
      "height": 139
    },
    ...
  ]
}

objects

An example of looks as follows.

{
  "image": <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=800x600 at 0x7F2F60698610>,
  "image_id": 1,
  "url": "https://cs.stanford.edu/people/rak248/VG_100K_2/1.jpg",
  "width": 800,
  "height": 600,
  "coco_id": null,
  "flickr_id": null,
  "objects": [
    {
      "object_id": 1058498,
      "x": 421,
      "y": 91,
      "w": 79,
      "h": 339,
      "names": [
        "clock"
      ],
      "synsets": [
        "clock.n.01"
      ]
    },
    ...
  ]
}

attributes

An example of looks as follows.

{
  "image": <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=800x600 at 0x7F2F60698610>,
  "image_id": 1,
  "url": "https://cs.stanford.edu/people/rak248/VG_100K_2/1.jpg",
  "width": 800,
  "height": 600,
  "coco_id": null,
  "flickr_id": null,
  "attributes": [
    {
      "object_id": 1058498,
      "x": 421,
      "y": 91,
      "w": 79,
      "h": 339,
      "names": [
        "clock"
      ],
      "synsets": [
        "clock.n.01"
      ],
      "attributes": [
        "green",
        "tall"
      ]
    },
    ...
  }
]

relationships

An example of looks as follows.

{
  "image": <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=800x600 at 0x7F2F60698610>,
  "image_id": 1,
  "url": "https://cs.stanford.edu/people/rak248/VG_100K_2/1.jpg",
  "width": 800,
  "height": 600,
  "coco_id": null,
  "flickr_id": null,
  "relationships": [
    {
      "relationship_id": 15927,
      "predicate": "ON",
      "synsets": "['along.r.01']",
      "subject": {
        "object_id": 5045,
        "x": 119,
        "y": 338,
        "w": 274,
        "h": 192,
        "names": [
          "shade"
        ],
        "synsets": [
          "shade.n.01"
        ]
      },
      "object": {
        "object_id": 5046,
        "x": 77,
        "y": 328,
        "w": 714,
        "h": 262,
        "names": [
          "street"
        ],
        "synsets": [
          "street.n.01"
        ]
      }
    }
    ...
  }
]

question_answers

An example of looks as follows.

{
  "image": <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=800x600 at 0x7F2F60698610>,
  "image_id": 1,
  "url": "https://cs.stanford.edu/people/rak248/VG_100K_2/1.jpg",
  "width": 800,
  "height": 600,
  "coco_id": null,
  "flickr_id": null,
  "qas": [
    {
      "qa_id": 986768,
      "image_id": 1,
      "question": "What color is the clock?",
      "answer": "Green.",
      "a_objects": [],
      "q_objects": []
    },
    ...
  }
]

Data Fields

When loading a specific configuration, users has to append a version dependent suffix:

from datasets import load_dataset
load_dataset("visual_genome", "region_description_v1.2.0")

region_descriptions

• image: A PIL.Image.Image object containing the image. Note that when accessing the image column: dataset[0]["image"] the image file is automatically decoded. Decoding of a large number of image files might take a significant amount of time. Thus it is important to first query the sample index before the "image" column, i.e. dataset[0]["image"] should always be preferred over dataset["image"][0]
• image_id: Unique numeric ID of the image.
• url: URL of source image.
• width: Image width.
• height: Image height.
• coco_id: Id mapping to MSCOCO indexing.
• flickr_id: Id mapping to Flicker indexing.
• regions: Holds a list of Region dataclasses:
  • region_id: Unique numeric ID of the region.
  • image_id: Unique numeric ID of the image.
  • x: x coordinate of bounding box's top left corner.
  • y: y coordinate of bounding box's top left corner.
  • width: Bounding box width.
  • height: Bounding box height.

objects

• image: A PIL.Image.Image object containing the image. Note that when accessing the image column: dataset[0]["image"] the image file is automatically decoded. Decoding of a large number of image files might take a significant amount of time. Thus it is important to first query the sample index before the "image" column, i.e. dataset[0]["image"] should always be preferred over dataset["image"][0]
• image_id: Unique numeric ID of the image.
• url: URL of source image.
• width: Image width.
• height: Image height.
• coco_id: Id mapping to MSCOCO indexing.
• flickr_id: Id mapping to Flicker indexing.
• objects: Holds a list of Object dataclasses:
  • object_id: Unique numeric ID of the object.
  • x: x coordinate of bounding box's top left corner.
  • y: y coordinate of bounding box's top left corner.
  • w: Bounding box width.
  • h: Bounding box height.
  • names: List of names associated with the object. This field can hold multiple values in the sense the multiple names are considered as acceptable. For example: ['monitor', 'computer'] at https://cs.stanford.edu/people/rak248/VG_100K/3.jpg
  • synsets: List of WordNet synsets.

attributes

• image: A PIL.Image.Image object containing the image. Note that when accessing the image column: dataset[0]["image"] the image file is automatically decoded. Decoding of a large number of image files might take a significant amount of time. Thus it is important to first query the sample index before the "image" column, i.e. dataset[0]["image"] should always be preferred over dataset["image"][0]
• image_id: Unique numeric ID of the image.
• url: URL of source image.
• width: Image width.
• height: Image height.
• coco_id: Id mapping to MSCOCO indexing.
• flickr_id: Id mapping to Flicker indexing.
• attributes: Holds a list of Object dataclasses:
  • object_id: Unique numeric ID of the region.
  • x: x coordinate of bounding box's top left corner.
  • y: y coordinate of bounding box's top left corner.
  • w: Bounding box width.
  • h: Bounding box height.
  • names: List of names associated with the object. This field can hold multiple values in the sense the multiple names are considered as acceptable. For example: ['monitor', 'computer'] at https://cs.stanford.edu/people/rak248/VG_100K/3.jpg
  • synsets: List of WordNet synsets.
  • attributes: List of attributes associated with the object.

relationships

• image: A PIL.Image.Image object containing the image. Note that when accessing the image column: dataset[0]["image"] the image file is automatically decoded. Decoding of a large number of image files might take a significant amount of time. Thus it is important to first query the sample index before the "image" column, i.e. dataset[0]["image"] should always be preferred over dataset["image"][0]
• image_id: Unique numeric ID of the image.
• url: URL of source image.
• width: Image width.
• height: Image height.
• coco_id: Id mapping to MSCOCO indexing.
• flickr_id: Id mapping to Flicker indexing.
• relationships: Holds a list of Relationship dataclasses:
  • relationship_id: Unique numeric ID of the object.
  • predicate: Predicate defining relationship between a subject and an object.
  • synsets: List of WordNet synsets.
  • subject: Object dataclass. See subsection on objects.
  • object: Object dataclass. See subsection on objects.

question_answers

• image: A PIL.Image.Image object containing the image. Note that when accessing the image column: dataset[0]["image"] the image file is automatically decoded. Decoding of a large number of image files might take a significant amount of time. Thus it is important to first query the sample index before the "image" column, i.e. dataset[0]["image"] should always be preferred over dataset["image"][0]
• image_id: Unique numeric ID of the image.
• url: URL of source image.
• width: Image width.
• height: Image height.
• coco_id: Id mapping to MSCOCO indexing.
• flickr_id: Id mapping to Flicker indexing.
• qas: Holds a list of Question-Answering dataclasses:
  • qa_id: Unique numeric ID of the question-answer pair.
  • image_id: Unique numeric ID of the image.
  • question: Question.
  • answer: Answer.
  • q_objects: List of object dataclass associated with question field. See subsection on objects.
  • a_objects: List of object dataclass associated with answer field. See subsection on objects.

Data Splits

All the data is contained in training set.

Dataset Creation

Curation Rationale

Source Data

Initial Data Collection and Normalization

Who are the source language producers?

Annotations

Annotation process

Who are the annotators?

From the paper:

We used Amazon Mechanical Turk (AMT) as our primary source of annotations. Overall, a total of over 33, 000 unique workers contributed to the dataset. The dataset was collected over the course of 6 months after 15 months of experimentation and iteration on the data representation. Approximately 800, 000 Human Intelligence Tasks (HITs) were launched on AMT, where each HIT involved creating descriptions, questions and answers, or region graphs. Each HIT was designed such that workers manage to earn anywhere between $6-$8 per hour if they work continuously, in line with ethical research standards on Mechanical Turk (Salehi et al., 2015). Visual Genome HITs achieved a 94.1% retention rate, meaning that 94.1% of workers who completed one of our tasks went ahead to do more. [...] 93.02% of workers contributed from the United States. The majority of our workers were between the ages of 25 and 34 years old. Our youngest contributor was 18 years and the oldest was 68 years old. We also had a near-balanced split of 54.15% male and 45.85% female workers.

Personal and Sensitive Information

Considerations for Using the Data

Social Impact of Dataset

Discussion of Biases

Other Known Limitations

Additional Information

Dataset Curators

Licensing Information

Visual Genome by Ranjay Krishna is licensed under a Creative Commons Attribution 4.0 International License.

Citation Information

@article{Krishna2016VisualGC,
  title={Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations},
  author={Ranjay Krishna and Yuke Zhu and Oliver Groth and Justin Johnson and Kenji Hata and Joshua Kravitz and Stephanie Chen and Yannis Kalantidis and Li-Jia Li and David A. Shamma and Michael S. Bernstein and Li Fei-Fei},
  journal={International Journal of Computer Vision},
  year={2017},
  volume={123},
  pages={32-73},
  url={https://doi.org/10.1007/s11263-016-0981-7},
  doi={10.1007/s11263-016-0981-7}
}

Contributions

Due to limitation of the dummy_data creation, we provide a fix_generated_dummy_data.py script that fix the dataset in-place.

Thanks to @thomasw21 for adding this dataset.