AWA2

Overview

Animals with Attributes 2 (AWA2) is an image classification dataset featuring 50 animal classes, primarily used for attribute-based image recognition and zero-shot learning research. The dataset provides a rich collection of animal images across diverse species.

Test: 37,322 images across 50 animal classes
Train: N/A (test-only dataset)

The dataset is widely used for zero-shot learning tasks where models must recognize animal classes they haven’t been trained on, using semantic attributes as a bridge between seen and unseen classes.

Data Structure

When accessing an example using ds[i], you will receive a dictionary with the following keys:

Key	Type	Description
`image`	`PIL.Image.Image`	RGB image (variable dimensions)
`label`	int	Class label (0-49)

Animal Classes

The dataset includes 50 animal classes:

antelope, grizzly+bear, killer+whale, beaver, dalmatian, persian+cat, horse, german+shepherd, blue+whale, siamese+cat, skunk, mole, tiger, hippopotamus, leopard, moose, spider+monkey, humpback+whale, elephant, gorilla, ox, fox, sheep, seal, chimpanzee, hamster, squirrel, rhinoceros, rabbit, bat, giraffe, wolf, chihuahua, rat, weasel, otter, buffalo, zebra, giant+panda, deer, bobcat, pig, lion, mouse, polar+bear, collie, walrus, raccoon, cow, dolphin

Usage Example

Basic Usage

from stable_datasets.images.awa2 import AWA2

# Load the test set (only split available)
ds = AWA2(split="test")

sample = ds[0]
print(sample.keys())  # {"image", "label"}
print(f"Label: {sample['label']}")  # Integer label (0-49)

# Access the image
image = sample["image"]
print(f"Image size: {image.size}")  # Variable dimensions

# Optional: make it PyTorch-friendly
ds_torch = ds.with_format("torch")

Getting Class Names

from stable_datasets.images.awa2 import AWA2

ds = AWA2(split="test")

# Get class names
class_names = ds.info.features["label"].names

# Map label to class name
sample = ds[0]
class_name = class_names[sample["label"]]
print(f"Animal: {class_name}")

Typical Use Case: Zero-Shot Learning

from stable_datasets.images.awa2 import AWA2

ds = AWA2(split="test")

# Define seen and unseen classes for zero-shot learning
seen_classes = [0, 1, 2, 3, 4]  # Train on these
unseen_classes = [5, 6, 7, 8, 9]  # Test on these

# Filter dataset
seen_data = ds.filter(lambda x: x["label"] in seen_classes)
unseen_data = ds.filter(lambda x: x["label"] in unseen_classes)

References

Homepage: https://cvml.ista.ac.at/AwA2/
Paper: Zero-Shot Learning—A Comprehensive Evaluation of the Good, the Bad and the Ugly (IEEE TPAMI 2019)

Citation

@ARTICLE{8413121,
  author={Xian, Yongqin and Lampert, Christoph H. and Schiele, Bernt and Akata, Zeynep},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  title={Zero-Shot Learning—A Comprehensive Evaluation of the Good, the Bad and the Ugly},
  year={2019},
  volume={41},
  number={9},
  pages={2251-2265},
  keywords={Semantics;Visualization;Task analysis;Training;Fish;Protocols;Learning systems;Generalized zero-shot learning;transductive learning;image classification;weakly-supervised learning},
  doi={10.1109/TPAMI.2018.2857768}