MAY199/synthetic-sofa-images

Synthetic Sofa Interior Images Dataset

Dataset Description

This dataset contains 1,000 synthetic images of sofas in indoor living room environments, generated using a text-to-image diffusion model.
The dataset was created as part of an AI project focused on visual similarity, embeddings, and recommendation systems.

The images depict a wide range of sofa styles, colors, and interior design aesthetics, and are intended for use in computer vision tasks such as image embeddings, similarity search, and recommendation pipelines.

---

Data Generation

The images were synthetically generated using a Stable Diffusion XL (SDXL) model with carefully designed prompts describing sofas in various interior styles (e.g., Scandinavian, Industrial, Classic, Modern).

Generation parameters were chosen to balance image quality and diversity:

• Resolution: 512×512
• Diverse prompts and color variations
• Negative prompts to reduce artifacts and unrealistic outputs

The full data generation process is documented in the provided notebook:

• synthetic_data_generation.ipynb

---

Exploratory Data Analysis (EDA)

Exploratory Data Analysis was conducted to assess data quality and consistency.

Dataset Overview

• Total images: 1,000
• File format: JPG
• Naming convention: sequential (sofa_0000.jpg to sofa_0999.jpg)

Visual Inspection

• Random samples of images were visually inspected.
• The majority of images depict realistic sofas in indoor living room scenes.
• A small number of visually atypical images were observed but retained to avoid introducing subjective bias.

Automated Integrity Checks

• All images were successfully loaded without corruption.
• No missing or unreadable files were found.
• All images share a consistent resolution of 512×512 pixels.

Brightness Distribution

• Image brightness distribution shows a balanced range of lighting conditions.
• No extreme bias toward overly dark or overly bright images was detected.

Based on the EDA, the dataset was deemed clean and suitable for downstream embedding extraction and similarity-based tasks.

The full EDA process and outputs are documented in:

• eda.ipynb

---

Intended Use

This dataset is intended for:

• Image embedding extraction
• Visual similarity search
• Recommendation systems
• Research and educational purposes

It is especially suitable for projects involving multimodal or vision-based representation learning.

---

Limitations

• The dataset is fully synthetic and may not capture all nuances of real-world photography.
• No explicit labels (e.g., style categories) are provided.
• Visual diversity depends on prompt design and generation settings.

---

Files in This Repository

• sofa_dataset_images/ – Image files
• eda.ipynb – Exploratory Data Analysis notebook
• synthetic_data_generation.ipynb – Synthetic data generation notebook
• README.md – Dataset documentation

---