EcomFruitAI

Creation of Synthetic Fruit Images with Diffusion Models

EcomFruitAI is a deep learning project that generates synthetic fruit images using diffusion models. Trained on the Fruits-360 dataset, this system can create fruit images from text descriptions, enabling applications in e-commerce, computer vision research, and data augmentation.

Features

• Text-to-Image Generation: Generate realistic fruit images from natural language descriptions
• Pre-trained Models: Uses CLIP text encoder and Stable Diffusion VAE components
• Custom UNet Architecture: Optimized for fruit image generation
• Modular Design: Clean, organized codebase for easy extension and maintenance
• Google Colab Compatible: Designed to run efficiently in cloud environments

Quick Start

Installation

1. Clone the repository:

   git clone https://github.com/ISCODEVUTB/EcomFruitAI
   cd EcomFruitAI

2. Install dependencies:

   pip install -r requirements.txt

3. Install the package in development mode:

   pip install -e .

Basic Usage

Using the Jupyter Notebook (Recommended)

Open and run the main notebook for a complete walkthrough:

jupyter lab notebooks/ecomfruitai.ipynb

Training Configuration

Key training parameters can be modified in ecomfruitai/config.py:

TRAINING_CONFIG = {
    "learning_rate": 1e-4,
    "num_epochs": 2,
    "batch_size": 16,
    "gradient_accumulation_steps": 2,
    "subset_size": 1000,  # For faster training
    "checkpoint_frequency": 100,
    "test_generation_frequency": 50
}

Generation Examples

Single Image Generation

from ecomfruitai.modeling.predict import generate_image
from ecomfruitai.plots import show_generated_image

# Generate image
image = generate_image("green apple, whole fruit, realistic photo", models)

# Display
show_generated_image(image, title="Generated Green Apple")

Batch Generation

from ecomfruitai.modeling.predict import generate_multiple_images
from ecomfruitai.plots import show_multiple_generated_images

# Define prompts
prompts = [
    "red apple, whole fruit, realistic photo",
    "yellow banana, whole fruit, realistic photo",
    "orange carrot, whole vegetable, realistic photo"
]

# Generate multiple images
images = generate_multiple_images(prompts, models)

# Display grid
show_multiple_generated_images(images, prompts)

Notebooks visualization

Notebooks

Dataset

The project uses the Fruits-360 dataset from Kaggle, which contains:

• 137,104 total images of fruits, vegetables, nuts and seeds
• 201 classes (fruits, vegetables, nuts and seeds)
• 100x100 pixel resolution
• Training set: 102,790 images
• Test set: 34,314 images

The system automatically filters classes with descriptive information (colors, varieties, conditions) for better text-to-image alignment.

Configuration

All project settings are centralized in ecomfruitai/config.py:

• Model configurations: Architecture parameters, pre-trained model paths
• Training settings: Learning rates, batch sizes, checkpointing
• Data processing: Image transforms, normalization parameters
• Generation parameters: Inference steps, sampling configurations

Project Organization

├── LICENSE            <- Open-source license if one is chosen
├── Makefile           <- Makefile with convenience commands like `make data` or `make train`
├── README.md          <- The top-level README for developers using this project.
├── data
│   ├── external       <- Data from third party sources.
│   ├── interim        <- Intermediate data that has been transformed.
│   ├── processed      <- The final, canonical data sets for modeling.
│   └── raw            <- The original, immutable data dump.
│
├── docs               <- A default mkdocs project; see www.mkdocs.org for details
│
├── models             <- Trained and serialized models, model predictions, or model summaries
│
├── notebooks          <- Jupyter notebooks. Naming convention is a number (for ordering),
│                         the creator's initials, and a short `-` delimited description, e.g.
│                         `1.0-jqp-initial-data-exploration`.
│
├── pyproject.toml     <- Project configuration file with package metadata for
│                         ecomfruitai and configuration for tools like black
│
├── references         <- Data dictionaries, manuals, and all other explanatory materials.
│
├── reports            <- Generated analysis as HTML, PDF, LaTeX, etc.
│   └── figures        <- Generated graphics and figures to be used in reporting
│
├── requirements.txt   <- The requirements file for reproducing the analysis environment, e.g.
│                         generated with `pip freeze > requirements.txt`
│
│
└── ecomfruitai   <- Source code for use in this project.
    │
    ├── __init__.py             <- Makes ecomfruitai a Python module
    │
    ├── config.py               <- Store useful variables and configuration
    │
    ├── dataset.py              <- Scripts to download or generate data
    │
    ├── features.py             <- Code to create features for modeling
    │
    ├── modeling
    │   ├── __init__.py
    │   ├── predict.py          <- Code to run model inference with trained models
    │   └── train.py            <- Code to train models
    │
    └── plots.py                <- Code to create visualizations

Acknowledgments

• Fruits-360 Dataset by Mihai Oltean
• Hugging Face Diffusers library
• Stability AI for VAE components
• OpenAI CLIP for text encoding

---