metadata

language:
  - en
pretty_name: Rubber Duck Detection Dataset
tags:
  - computer-vision
  - object-detection
  - image-classification
  - rubber-ducks
  - yolo
annotations_creators:
  - expert-generated
  - machine-generated
language_creators:
  - found
size_categories:
  - n<1K
source_datasets:
  - Norod78/Rubber-Duck-blip-captions
  - linoyts/rubber_ducks
task_categories:
  - object-detection
  - image-classification
task_ids:
  - multi-class-image-classification
dataset_info:
  features:
    - name: image_path
      dtype: string
    - name: detections
      sequence: string
    - name: verified
      dtype: 'null'
  splits:
    - name: default
      num_bytes: 23643
      num_examples: 192
  download_size: 15317
  dataset_size: 23643
  configs:
    - config_name: default
      path:
        - data/*.jpg
        - data/*.txt
        - train-00000-of-00001.parquet

Rubber Duck Detection Dataset

Overview

This dataset contains 192 annotated images of rubber ducks, specifically curated for object detection tasks. It was used for experimentation related to the YOLOv8n Rubber Duck Detector model.

NOTE: I DO NOT RECOMMEND USING THIS DATASET AT THIS TIME. There is an open and ongoing discussion around the use of the datasets that were combined for this.
See related licensing discussion on the forum

Dataset Description

Dataset Summary

A specialized computer vision dataset featuring rubber ducks with bounding box annotations, designed for training object detection models. The dataset combines and enhances images from two existing datasets with manual verification and standardized annotations.

Supported Tasks

Object Detection
Image Classification

Languages

Visual data with English annotations

Dataset Structure

Number of examples: 192
Download size: 15,317 bytes
Dataset size: 23,643 bytes

Data Fields

image_path: Path to the image file
detections: Bounding box coordinates in YOLO format (normalized)
verified: Verification status field

Data Splits

Training set: 192 images

Source Data

Initial Data Collection and Normalization

This dataset combines images from two existing datasets:

Norod78/Rubber-Duck-blip-captions
linoyts/rubber_ducks NOTE: these datasets were unlicensed so usage is not clear. See related licensing discussion on the forum

Data Format

Each image has corresponding annotations in two formats:

YOLO text format (.txt files)
Structured format in the parquet file

The YOLO format annotations follow the convention:

All values are normalized between 0 and 1
Single class: rubber duck
Coordinates represent the center point and dimensions of the bounding box

File Structure

data/ ├── images/ │ ├── image_0.jpg │ ├── image_1.jpg │ └── ... ├── labels/ │ ├── image_0.txt │ ├── image_1.txt │ └── ... └── train-00000-of-00001.parquet Copy

Dataset Creation

Curation Rationale

This dataset was created to provide a specialized dataset for rubber duck detection, combining images from existing datasets with standardized annotations. The purpose was to see if we could enhance YOLOv8n to be better at detecting an assortment of rubber ducks.

Annotations

Annotations were created using a combination of:

Manual annotation
Machine-generated predictions with human verification
Standardization to YOLO format

Usage

Loading the Dataset

from datasets import load_dataset

dataset = load_dataset("brainwavecollective/yolo-rubber-ducks")

Considerations for Using the Data

Discussion of Biases

Dataset focuses specifically on strange rubber ducks
Does not represent all possible variations of rubber ducks
Limited environmental contexts
Most of the images have a marketing focus

Additional Information

Dataset Curators

This dataset was curated by combining and enhancing existing rubber duck datasets with additional annotations and verification.

Licensing Information

NOTE: Licensing is currently under review. See related licensing discussion on the forum

Citation Information

If you use this dataset, please cite:

@misc{rubber-duck-detection,
  author = {Daniel Ritchie},
  title = {Rubber Duck Detection Dataset},
  year = {2025},
  publisher = {HuggingFace},
  journal = {HuggingFace Hub},
  howpublished = {\url{https://huggingface.co/datasets/brainwavecollective/yolo-rubber-ducks}}
}