Data Framework for Image Inpainting

A sophisticated framework for generating dynamic masks and performing image inpainting using state-of-the-art diffusion models. This framework automates the process of creating diverse mask patterns and applying intelligent inpainting for image manipulation.

Features

• Dynamic Mask Generation: Creates various types of masks including:

  • Rectangle shapes
  • Circular patterns
  • Elliptical shapes
  • Bacteria-like organic patterns
  • Freeform shapes with natural contours

• Automated Inpainting Pipeline: Utilizes advanced diffusion models for seamless image restoration

• Configurable Parameters: Easily customize mask generation and inpainting settings

• COCO Dataset Integration: Built-in support for the COCO dataset

• Logging System: Comprehensive logging for pipeline monitoring

Installation

1. Clone the repository:

   git clone https://github.com/raghulchandramouli/data_framework
   cd data_framework

2. Install the required dependencies:

   pip install -r requirements.txt

Configuration

The framework uses a YAML configuration file (config.yaml) to manage various settings:

paths:
  coco_dir: "path/to/coco/dataset"
  mask_dir: "path/to/mask/output"
  inpainted_dir: "path/to/inpainted/output"

dataset:
  num_images_to_process: 100
  start_index: 0

mask:
  dilation_kernel: 15
  dilation_iterations: 5

Usage

Running the Pipeline

Command Line Execution

You can run the pipeline directly from the command line:

python pipeline.py

To specify a custom configuration file:

python pipeline.py custom_config.yaml

Python Script Execution

Alternatively, you can import and run the pipeline in a Python script:

from pipeline import run_pipeline

# Run with default configuration
run_pipeline()

# Or specify a custom config file
run_pipeline("custom_config.yaml")

Mask Generation

from mask_generator import generate_dynamic_mask

generate_dynamic_mask(
    image_path="path/to/input/image.jpg",
    mask_path="path/to/output/mask.png"
)

Inpainting

from inpainting import inpaint_image

inpaint_image(
    pipe=your_inpainting_pipeline,
    image_path="path/to/input/image.jpg",
    mask_path="path/to/mask.png",
    output_path="path/to/output/inpainted.jpg"
)

Components

Mask Generator

The mask generator creates dynamic binary masks using various geometric shapes:

• Rectangle: Creates rectangular masks with random positions and sizes
• Circle: Generates circular masks with varying radii
• Ellipse: Produces elliptical masks with random orientations
• Bacteria: Creates organic, bacteria-like shapes using random points
• Freeform: Generates natural-looking irregular shapes

Inpainting Pipeline

The inpainting system uses a state-of-the-art diffusion model to fill masked regions with contextually appropriate content. Key features include:

• Customizable prompt for guiding the inpainting process
• Adjustable strength parameter for controlling the intensity of changes
• Configurable number of inference steps for quality control

Project Structure

├── config.py           # Configuration loading utilities
├── config.yaml         # Configuration settings
├── dataset_manager.py  # Dataset handling functions
├── inpainting.py       # Inpainting implementation
├── mask_generator.py   # Mask generation logic
├── pipeline.py         # Main pipeline orchestration
├── requirements.txt    # Project dependencies
└── utils.py           # Utility functions

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

[MIT License]