Denoising Framework

A robust framework for image denoising and defect detection, leveraging a multi-phase deep learning approach to effectively handle various noise types and enhance defect detection accuracy.

Features

Phase 1: Defect Detection

• Implements a binary classification model for accurate defect detection
• Utilizes BCEWithLogitsLoss for model training
• Employs AdamW optimizer with configurable learning rate

Phase 2.1: Noise Type Detection

• Performs multi-class classification to identify noise types
• Supports Gaussian, Periodic, and Salt noise types
• Uses CrossEntropyLoss for training

Phase 2.2: Denoising Models

• Provides dedicated denoising models for each noise type
• Features a custom architecture optimized for noise removal
• Includes individual training pipelines for each noise type

Phase 3.1: Dataset Denoising

• Applies trained models to denoise the entire dataset
• Integrates noise type detection and denoising for seamless processing

Phase 3.2: Defect Detection on Denoised Data

• Retrains the defect detection model using denoised images

Installation

Follow these steps to set up the framework:

1. Clone the Repository:

   git clone https://github.com/safinal/denoising-framework.git
   cd denoising-framework

2. Set Up a Virtual Environment:

   python -m venv .venv
   source .venv/bin/activate  # On Linux/Mac
   # or
   .venv\Scripts\activate     # On Windows

3. Install Dependencies:

   pip install -r requirements.txt

Usage

Run the framework in different phases using the main script:

python run.py --phase [PHASE] --config [YAML_CONFIG_PATH]

Available Phases

• phase1: Trains the defect detection model
• phase2.1: Trains the noise type detection model
• phase2.2: Trains denoising models for each noise type
• phase3.1: Denoises the dataset using trained models
• phase3.2: Retrains defect detection on the denoised dataset

Configuration

The framework uses YAML configuration files located in the src/config/ directory to specify parameters.

License

This project is licensed under the terms of the included LICENSE file.