📖 API Reference

The Picsellia CV Engine API is a modular toolkit for building end-to-end pipelines. It’s organized into reusable components, decorators, and framework-specific extensions.

---

Core concepts

• Steps: Modular units of logic (e.g. load, train, validate)
• Pipelines: Logical flows decorated with @pipeline, composed of @step
• Contexts: Injected objects carrying pipeline configuration and metadata

---

Built-in components

Base steps

Dataset

• Loader
• Preprocessor
• Uploader
• Validator

Model

• Builder
• Evaluator

Datalake

• Loader

---

Framework-Specific extensions

Frameworks are isolated under: src/picsellia_cv_engine/frameworks/<framework_name>/

Each framework can include:

• Custom model modules
• Hyperparameter definitions
• Training, evaluation, or export services
• Framework-specific steps

---

Decorators

• @pipeline – Defines a pipeline entrypoint
• @step – Marks a function as a step
• Step Metadata

---

Data models

• COCO Dataset
• Dataset Collection
• Datalake Collection

---

Code example

from picsellia_cv_engine.decorators.pipeline_decorator import pipeline
from picsellia_cv_engine.steps.base.dataset.loader import load_yolo_datasets

@pipeline
def my_pipeline():
    dataset = load_yolo_datasets()
    ...