Datasetpublic

srivathsanb14/dustbin-or-not-dataset

Dustbin vs. Not Dustbin Image Dataset Dataset Summary A small binary image classification dataset for predicting whether an image contains a dustbin/trash can. has_dustbin = 1: dustbin visible has_dustbin = 0: no dustbin visible 30+ original photographs Images resized to 224 × 224 RGB Data Collection Original photographs were captured by the author and organized into dustbin/ and not_dustbin/ folders. No people, faces, or sensitive personal… See the full description on the dataset page: https://huggingface.co/datasets/srivathsanb14/dustbin-or-not-dataset.

sourceHugging Facecc-by-4.0updated 9d agoView on Hugging Face

0likes45downloads

Dataset Card

Dustbin vs. Not Dustbin Image Dataset

Dataset Summary

A small binary image classification dataset for predicting whether an image contains a dustbin/trash can.

—has_dustbin = 1: dustbin visible
—has_dustbin = 0: no dustbin visible
—30+ original photographs
—Images resized to 224 × 224 RGB

Data Collection

Original photographs were captured by the author and organized into dustbin/ and not_dustbin/ folders. No people, faces, or sensitive personal information are intentionally included.

Preprocessing & Augmentation

Original images are split into approximately 70% train, 15% validation, and 15% test using stratification by label.

The training split is augmented with 300 synthetic images using label-preserving transformations:

—Rotation (±20°)
—Horizontal flip
—Brightness and contrast adjustments
—Gaussian noise
—Random zoom/crop

Augmentation is applied only to training images. Validation and test images remain unchanged.

Splits

Split	Contents
`train`	~70% original + 300 augmented
`validation`	~15% original
`test`	~15% original

Intended Use

Educational use for practicing image classification, dataset construction, augmentation, and machine-learning workflows.

Limitations

The dataset is small and reflects a limited range of environments, cameras, lighting conditions, and dustbin types. Validation and test metrics may therefore be noisy, and models trained on this dataset may have limited real-world generalization.

License

CC BY 4.0, all photographs are original and contain no third-party copyrighted material.

AI Usage Disclosure

AI assistance (Claude, Anthropic) was used to develop the notebook, including data loading, preprocessing, splitting, augmentation, and Hugging Face dataset construction. The real photographs and their labels were supplied by the author. The optional demo-image generator is not part of the final dataset.