graph LR
CLI_Interface["CLI Interface"]
Training_Orchestrator["Training Orchestrator"]
Data_Pipeline["Data Pipeline"]
GAN_Model_Core["GAN Model Core"]
Evaluation_Persistence["Evaluation & Persistence"]
Experiment_Tracking["Experiment Tracking"]
CLI_Interface -- "initiates" --> Training_Orchestrator
Training_Orchestrator -- "requests data from" --> Data_Pipeline
Data_Pipeline -- "provides augmented data to" --> GAN_Model_Core
Training_Orchestrator -- "trains" --> GAN_Model_Core
Training_Orchestrator -- "interacts with" --> Evaluation_Persistence
Evaluation_Persistence -- "utilizes" --> GAN_Model_Core
Training_Orchestrator -- "sends metrics to" --> Experiment_Tracking
click Training_Orchestrator href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/lightweight-gan/Training_Orchestrator.md" "Details"
click Data_Pipeline href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/lightweight-gan/Data_Pipeline.md" "Details"
click Evaluation_Persistence href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/lightweight-gan/Evaluation_Persistence.md" "Details"
The lightweight-gan architecture is designed as a streamlined ML toolkit, centered around a Training Orchestrator that drives the entire GAN training and evaluation process. User interaction begins with the CLI Interface, which configures and launches the Training Orchestrator. Data flows through a Data Pipeline that handles loading and augmentation before being fed into the GAN Model Core, where the Generator and Discriminator models learn. The Training Orchestrator periodically leverages the Evaluation & Persistence component for performance assessment, synthetic image generation, and checkpoint management. All critical training metrics and progress are logged via the Experiment Tracking component, providing comprehensive insights into experiment runs. This modular design facilitates clear data and control flow, making it ideal for visual representation in a flow graph diagram.
Entry point for user interaction, parsing commands and configurations.
Related Classes/Methods:
Training Orchestrator [Expand]
Manages the entire GAN training lifecycle, coordinating data flow and component interactions.
Related Classes/Methods:
Data Pipeline [Expand]
Handles loading, preprocessing, and augmenting image datasets for the GAN.
Related Classes/Methods:
Encapsulates the Generator and Discriminator neural networks, responsible for image generation and discrimination.
Related Classes/Methods:
Evaluation & Persistence [Expand]
Manages model performance evaluation, synthetic image generation, and saving/loading model states.
Related Classes/Methods:
Integrates with Aim for logging and monitoring training experiments.
Related Classes/Methods: