Capstone-SLAC

A deep learning framework for image classification using PyTorch, designed for the SLAC project.

Project Structure

Data

• Contains training, testing, and validation CSV files with image paths
• Organized to facilitate easy data loading and processing

Images

• Sample images showing original and transformed versions
• Useful for visualizing preprocessing techniques

Models

• Saved model checkpoints and performance metrics
• JSON files from model testing
• Performance visualization plots (PNG)

SLAC25 Package

Core functionality organized into modular components:

Data Handling

• dataloader.py: Efficient batch loading with customizable sampling strategies

  • DataLoaderFactory class for creating and configuring data loaders
  • Various sampling methods (sequential, random, weighted)
  • Optimized for performance with multi-worker support

• dataset.py: Custom dataset implementation

  • Image loading and transformation pipeline
  • Efficient memory management for large datasets

Model Architecture

• models.py: Neural network model definitions
  • BaselineCNN: Simple convolutional neural network
  • ResNet: Residual network implementation
  • Easily extensible for new architectures

Training Infrastructure

• network.py: Training and evaluation framework
  • Wrapper: Base class for model training setup
  • ModelWrapper: High-level interface for model training, validation, and testing
  • Support for test mode with reduced dataset size for quick iterations

Utilities

• sampler.py: Custom sampling strategies
• transform.py: Image transformation and augmentation
• utils.py: Helper functions and classes
  • Model evaluation metrics
  • Performance visualization
  • Early stopping implementation
  • Additional utilities

Main Script

• main.py: Entry point with command-line interface
  • Flexible argument parsing for training configuration
  • Easy model selection and hyperparameter tuning
  • Test mode for rapid prototyping and debugging

Usage

Basic usage:

python __main__.py --nepoch 10 --batch_size 32

Test mode (for quick iterations):

python __main__.py --testmode

Features

• Modular Design: Easily swap models, datasets, and training strategies
• Test Mode: Quickly validate code changes with smaller datasets
• Comprehensive Logging: Detailed training metrics and model checkpoints
• Early Stopping: Prevent overfitting with validation-based early stopping
• Learning Rate Scheduling: Adaptive learning rate for improved convergence