Skip to content

api-reference.md

Latest commit

 

History

History
74 lines (46 loc) · 3.07 KB

api-reference.md

File metadata and controls

74 lines (46 loc) · 3.07 KB
Raw

API Reference

This document describes the main types and interfaces of this C++23 project. Implementations will live in the corresponding headers and source files.

Core Module (normitri::core)

Image and frame types

  • Frame — Represents a single image or video frame.

    • Dimensions (width, height), channel count, element type (e.g. uint8_t, float).
    • Buffer storage (owned or non-owning view); access via std::span or raw pointer where needed for interop with CV/ML APIs.

  • PixelFormat — Enum or type describing layout (e.g. RGB, BGR, grayscale, planar vs packed).

Error handling

  • PipelineError (or equivalent) — Enum or variant of error codes:
    • e.g. InvalidFrame, LoadFailed, InferenceFailed, InvalidConfig.
  • std::expected<T, E> — Return type for operations that can fail; E is typically PipelineError or a small error type.

Pipeline contract

  • IPipelineStage (or concept PipelineStage) — Contract for a single stage:

    • process(Frame const& in) -> std::expected<Frame, PipelineError> (or similar).
    • Stages are composable; the pipeline invokes them in sequence.

  • Pipeline — Holds an ordered list of stages; run(Frame const&) -> std::expected<Result, PipelineError>.

Vision Module (normitri::vision)

Preprocessing

  • ResizeStage — Resizes input frame to a given width/height (e.g. for inference input size).
  • NormalizeStage — Applies mean/scale or similar normalization (e.g. for neural network input).
  • ColorConvertStage — Converts between pixel formats (e.g. BGR → RGB, grayscale).

Parameters (dimensions, normalization constants) are configurable via constructor or a config struct.

Inference adapter

  • IInferenceBackend (or concept) — Abstract interface for running inference:
    • infer(Frame const& input) -> std::expected<InferenceResult, PipelineError>.
  • Concrete backends (ONNX, TensorRT, etc.) implement this interface so the pipeline stays backend-agnostic.

Results

  • InferenceResult — Holds raw or decoded outputs (e.g. detection boxes, class scores, segmentation mask).
  • Detection — Single detection: bounding box, class id, confidence (used when decoding detection outputs).

App Module

  • main() — Parses CLI (e.g. input path, config path), constructs pipeline from config, runs on one or more frames, and prints or logs results.
  • PipelineRunner (optional) — Wraps pipeline execution with threading (e.g. thread pool or async) for batch or stream processing.

Namespaces Summary

Namespace Purpose
normitri Top-level; re-exports or aggregates public API.
normitri::core Frame, pipeline contract, errors.
normitri::vision Preprocessing stages, inference adapter, result types.
normitri::detail Implementation details; not part of stable API.

This API reference will be updated as types and function signatures are added to the codebase. For build and usage, see Building and Architecture.