metadata

license: mit
task_categories:
  - image-classification
language:
  - en
tags:
  - biology
  - birds
  - fine-grained image classification
  - natural language description
size_categories:
  - 1K<n<10K

Dataset Card for CUB_200_2011

Dataset Description

Homepage: https://www.vision.caltech.edu/datasets/cub_200_2011/
Citation: @techreport{WahCUB_200_2011, Title = , Author = {Wah, C. and Branson, S. and Welinder, P. and Perona, P. and Belongie, S.}, Year = {2011} Institution = {California Institute of Technology}, Number = {CNS-TR-2011-001} }

Dataset Summary

The Caltech-UCSD Birds 200-2011 dataset (CUB-200-2011) is an extended version of the original CUB-200 dataset, featuring photos of 200 bird species primarily from North America. This 2011 version significantly expands its predecessor by doubling the number of images per class and introducing new part location annotations, alongside collecting detailed natural language descriptions for each image through Amazon Mechanical Turk (AMT). The dataset includes a total of 11,788 images, split into 5,994 for training and 5,794 for testing.

Supported Tasks and Leaderboards

This dataset can support a variety of computer vision tasks, including but not limited to:

Fine-Grained Image Classification
Object Detection and Localization
Semantic Segmentation
Attribute-Based Recognition
Multitask Learning

Languages

The dataset includes annotations in English

Dataset Structure

Data Instances

A data instance in the CUB-200-2011 dataset comprises an image of a bird species, along with annotations including bounding boxes, part locations, binary attributes, and natural language descriptions.

{
"text": "A photo of a Tropical King Bird",
"image": cassiekang/train-00000-of-00001-246c29c8515f0b3f/Tropical_Kingbird_0064_69889.jpg
}

Data Fields

images: Photographs of birds across 200 species.
annotations: This includes:
- bounding boxes: Specify the bird's location within the image.
- segmentation labels: Provide pixel-wise segmentation for precise object segmentation.
- part locations: 15 specific parts of the bird are annotated for detailed analysis.
- binary attributes: 312 attributes indicating the presence or absence of certain features or behaviors.
- natural language descriptions: Ten single-sentence descriptions per image, collected via AMT.

Data Splits

Training set: 5,994 images
Test set: 5,794 images

Considerations for Using the Data

Social Impact of Dataset

The dataset contributes to advancements in computer vision, particularly in fine-grained image classification and object detection, with potential applications in biodiversity monitoring and species conservation.