SebastianAndreu/2025-24679-HW1-images

Dataset Card for Chair Presence Image Dataset

This dataset consists of original and augmented images labeled to indicate whether a chair is present in the scene (1) or not (0). It was built as part of a coursework project on dataset creation and augmentation.

Dataset Details

Dataset Description

The dataset contains 30 original images and 300 augmented images. Each sample is resized to 224x224 pixels.

• Labels: Binary (0 = no chair, 1 = chair present).

• Splits: original and augmented.

• Curated by: Sebastian Andreu (Carnegie Mellon University, coursework project)

• License: MIT

Uses

Direct Use

• Training and evaluating binary classifiers to detect the presence of a chair.
• Practicing data augmentation and dataset handling with Hugging Face Datasets.

Out-of-Scope Use

• Not suitable for object detection or segmentation.
• Not representative of all chair types, lighting conditions, or environments.
• Should not be used in production or safety-critical applications.

Dataset Structure

• Features:

  • image: RGB image (224x224)
  • label: integer (0 = no chair, 1 = chair present)

• Splits:

  • original: 30 manually collected images
  • augmented: 300 synthetic images

Dataset Creation

Curation Rationale

Chairs were chosen as the target class because they are common, safe to capture, and provide clear positive/negative cases.

Source Data

Data Collection and Processing

• Images were captured manually by the curator.
• All images were resized to 224x224 pixels.
• Augmentation techniques included flipping, rotation, scaling, and color jittering.

Who are the source data producers?

All original images were produced by the curator. Augmented images were generated algorithmically.

Annotations

Annotation process

Labels were applied manually by the curator.

Who are the annotators?

Annotations were made by the dataset curator.

Personal and Sensitive Information

No personal or sensitive information is included. No people or identifiable features appear in the dataset.

Bias, Risks, and Limitations

• The dataset is small and not diverse.
• Chairs are limited to a narrow set of scenes and styles.
• Generalization outside the dataset context will be weak.

Recommendations

This dataset should be treated as a toy dataset for experimentation and coursework.

Citation

BibTeX: