Datasets:
The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider
removing the
loading script
and relying on
automated data support
(you can use
convert_to_parquet
from the datasets
library). If this is not possible, please
open a discussion
for direct help.
Dataset Card for Visual Genome
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
: APIL.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 overdataset["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 ofRegion
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
: APIL.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 overdataset["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 ofObject
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.jpgsynsets
: List ofWordNet synsets
.
attributes
image
: APIL.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 overdataset["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 ofObject
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.jpgsynsets
: List ofWordNet synsets
.attributes
: List of attributes associated with the object.
relationships
image
: APIL.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 overdataset["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 ofRelationship
dataclasses:relationship_id
: Unique numeric ID of the object.predicate
: Predicate defining relationship between a subject and an object.synsets
: List ofWordNet synsets
.subject
: Object dataclass. See subsection onobjects
.object
: Object dataclass. See subsection onobjects
.
question_answers
image
: APIL.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 overdataset["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 ofQuestion-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 withquestion
field. See subsection onobjects
.a_objects
: List of object dataclass associated withanswer
field. See subsection onobjects
.
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.
- Downloads last month
- 655