nirmalendu01/animal_dataset

Animal Dataset from COCO

This dataset contains animal images extracted from the COCO 2017 training set, with two download configurations available.

Dataset Configurations

1. Metadata Configuration (Small Download)

The metadata configuration contains only text metadata and image URLs - no image bytes. This is ideal for quick exploration and when you want to download images yourself.

from datasets import load_dataset

# Load full metadata dataset (small download)
ds = load_dataset("nirmalendu01/animal_dataset", "metadata", split="train")

# Load first 200 samples only (partial download)
ds = load_dataset("nirmalendu01/animal_dataset", "metadata", split="train[:200]")

Fields included:

• cocoid: COCO image ID
• file_name: Image filename
• width, height: Image dimensions
• coco_url, flickr_url: Image download URLs
• captions: List of captions for the image
• animal_labels: List of animal types in the image
• animal_bboxes_xywh: Bounding boxes [[x, y, width, height], ...]
• animal_bbox_labels: Labels corresponding to each bbox

2. Images Configuration (Large Download)

The images configuration includes the actual image data. This is larger but convenient for direct use.

from datasets import load_dataset

# Load full images dataset (large download)
ds_img = load_dataset("nirmalendu01/animal_dataset", "images", split="train")

# Load first 200 samples only (partial download with images)
ds_img = load_dataset("nirmalendu01/animal_dataset", "images", split="train[:200]")

Fields included:

• All fields from metadata configuration
• image: PIL Image object with the actual image data

Partial Downloads and Slicing

This dataset supports partial downloads through split slicing. You can download only a subset of the data without downloading the entire dataset:

# First 100 samples
ds = load_dataset("nirmalendu01/animal_dataset", "metadata", split="train[:100]")

# Samples 100-200
ds = load_dataset("nirmalendu01/animal_dataset", "metadata", split="train[100:200]")

# Last 100 samples
ds = load_dataset("nirmalendu01/animal_dataset", "metadata", split="train[-100:]")

# Every 10th sample (10% of data)
ds = load_dataset("nirmalendu01/animal_dataset", "metadata", split="train[::10]")

Note: Split slicing works for both metadata and images configurations. The dataset is sharded into multiple files to enable efficient partial downloads.

Usage Examples

Example 1: Quick exploration with metadata only

from datasets import load_dataset

# Small download - just metadata
ds = load_dataset("nirmalendu01/animal_dataset", "metadata", split="train[:100]")

# Access metadata
print(ds[0]["animal_labels"])  # ['cat', 'dog']
print(ds[0]["captions"][0])    # "A cat and dog playing..."
print(ds[0]["coco_url"])       # Image URL

Example 2: Working with images

from datasets import load_dataset

# Larger download - includes images
ds = load_dataset("nirmalendu01/animal_dataset", "images", split="train[:100]")

# Access image directly
image = ds[0]["image"]
image.show()  # Display image

# Access metadata
print(ds[0]["animal_labels"])
print(ds[0]["animal_bboxes_xywh"])

Example 3: Object detection with bounding boxes

from datasets import load_dataset
from PIL import Image, ImageDraw

ds = load_dataset("nirmalendu01/animal_dataset", "images", split="train[:10]")

for example in ds:
    img = example["image"]
    bboxes = example["animal_bboxes_xywh"]  # [[x, y, w, h], ...]
    labels = example["animal_bbox_labels"]

    # Draw bounding boxes
    draw = ImageDraw.Draw(img)
    for bbox, label in zip(bboxes, labels):
        x, y, w, h = bbox
        draw.rectangle([x, y, x+w, y+h], outline="red", width=2)
        draw.text((x, y-10), label, fill="red")

    img.show()

Dataset Statistics

• Total images: 23,989
• Animal categories: 10 (bird, cat, dog, horse, sheep, cow, elephant, bear, zebra, giraffe)
• Source: COCO 2017 Training Set
• License: CC-BY-4.0 (inherited from COCO)

Citation

If you use this dataset, please cite the original COCO dataset:

@article{lin2014microsoft,
  title={Microsoft coco: Common objects in context},
  author={Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll{'a}r, Piotr and Zitnick, C Lawrence},
  journal={arXiv preprint arXiv:1405.0312},
  year={2014}
}