Skip to content

on_boarding.md

Latest commit

 

History

History
120 lines (76 loc) · 8.56 KB

on_boarding.md

File metadata and controls

120 lines (76 loc) · 8.56 KB
Raw
graph LR
    External_Systems["External Systems"]
    Core_Metric_API["Core Metric API"]
    Full_Reference_Metrics["Full-Reference Metrics"]
    No_Reference_Metrics["No-Reference Metrics"]
    Distribution_Based_Metrics["Distribution-Based Metrics"]
    Perceptual_Loss_Functions["Perceptual Loss Functions"]
    Feature_Extractors["Feature Extractors"]
    Image_Processing_Utilities["Image Processing Utilities"]
    External_Systems -- "Provides Input Images" --> Core_Metric_API
    External_Systems -- "Invokes Metrics" --> Core_Metric_API
    Full_Reference_Metrics -- "Provides Output" --> External_Systems
    No_Reference_Metrics -- "Provides Output" --> External_Systems
    Distribution_Based_Metrics -- "Provides Output" --> External_Systems
    Perceptual_Loss_Functions -- "Provides Output" --> External_Systems
    Core_Metric_API -- "Implemented By" --> Full_Reference_Metrics
    Core_Metric_API -- "Implemented By" --> No_Reference_Metrics
    Core_Metric_API -- "Implemented By" --> Distribution_Based_Metrics
    Core_Metric_API -- "Implemented By" --> Perceptual_Loss_Functions
    Full_Reference_Metrics -- "Uses for Preprocessing" --> Image_Processing_Utilities
    No_Reference_Metrics -- "Uses for Preprocessing" --> Image_Processing_Utilities
    Distribution_Based_Metrics -- "Utilizes Features From" --> Feature_Extractors
    Distribution_Based_Metrics -- "Uses for Preprocessing" --> Image_Processing_Utilities
    Perceptual_Loss_Functions -- "Utilizes Features From" --> Feature_Extractors
    click Full_Reference_Metrics href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/piq/Full_Reference_Metrics.md" "Details"
    click Distribution_Based_Metrics href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/piq/Distribution_Based_Metrics.md" "Details"
    click Perceptual_Loss_Functions href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/piq/Perceptual_Loss_Functions.md" "Details"
Loading

CodeBoardingDemoContact

Details

The piq library is designed to provide a comprehensive suite of image quality metrics. Its architecture is centered around a flexible Core Metric API that defines the common interface for all metrics. This API is implemented by various specialized metric components, including Full-Reference Metrics, No-Reference Metrics, Distribution-Based Metrics, and Perceptual Loss Functions, each addressing different evaluation paradigms. The system interacts with External Systems that provide input images and consume the computed quality scores. To support its diverse metric calculations, piq leverages Feature Extractors for high-level image representations and Image Processing Utilities for fundamental image manipulations. This modular design ensures extensibility and reusability across different image quality assessment tasks.

External Systems

Represents external sources of image data (e.g., datasets, applications) and frameworks that utilize piq for evaluation or training. This component is external to the piq library and interacts with it by providing input images and consuming metric outputs.

Related Classes/Methods: None

Core Metric API

The foundational interface defining common methods and abstract classes for all image quality metrics, ensuring a consistent interaction model. This component establishes the contract for how metrics are implemented and used.

Related Classes/Methods:

Full-Reference Metrics [Expand]

A collection of metrics (e.g., SSIM, MS-SSIM, FSIM) that require both a reference and a distorted image for comparison. These metrics inherit from the Core Metric API.

Related Classes/Methods:

No-Reference Metrics

Metrics (e.g., BRISQUE, Total Variation) that assess image quality without the need for an original reference image. These metrics also adhere to the Core Metric API.

Related Classes/Methods:

Distribution-Based Metrics [Expand]

Metrics (e.g., FID, IS, KID, PR) that evaluate image quality by comparing statistical distributions of features extracted from image sets. These metrics rely on Feature Extractors and the Core Metric API.

Related Classes/Methods:

Perceptual Loss Functions [Expand]

Metrics (e.g., PerceptualLoss) designed to be used as loss functions within neural network training, leveraging features from pre-trained models. These also utilize Feature Extractors and the Core Metric API.

Related Classes/Methods:

Feature Extractors

Dedicated modules for extracting high-level features from images using pre-trained deep learning models (e.g., InceptionV3, CLIP). These are crucial for distribution-based and perceptual loss metrics.

Related Classes/Methods:

Image Processing Utilities

A collection of low-level, functional utilities for common image manipulation tasks, including resizing and color space conversions, used by various metrics for preprocessing.

Related Classes/Methods:

FAQ