3. What do you want to achieve with Picsellia ?

3. What do you want to achieve with Picsellia ?

Objectives:

  • Understand the platform philosophy
  • Understand the platform structure
  • Understand what the Quick Start Guide will help your to achieve

1. Picsellia's philosophy

Picsellia's mission is to empower technical teams to upgrade their Computer Vision Models into Remarkable Computer Vision pipelines.

With Picsellia, AI teams can Structure, Build & Observe their CV pipelines.

  • Structure: Increase AI team time spent on high-value tasks by +33% to unleash collective intelligence.
  • Operate: Create & Deploy CV models 3X time faster, with scalable infrastructures and automation.
  • Observe: Gain actionable insights in seconds with full ML observability & experiment traceability.

You can leverage all the features on the Picsellia platform to achieve this goal.

2. Platform's structure

The platform is divided into 3 main parts:

  • Data Management (upload Data, create Dataset)
  • Data Science (train your Model and assess their quality)
  • Model Operations (Deploy, monitor, and retrain your Model automatically)

3. Purpose of the Quick Start Guide

This Quick Start Guide aims to help you gain a global understanding of the core capabilities of the Picsellia platform.

If you follow this guide, you will understand how to:

  • Import your images as Data to the Datalake
  • Create DatasetVersion
  • Annotate DatasetVersion and import an Annotation file
  • Fork a DatasetVersion
  • Train a ModelVersion by using a pre-trained architecture provided by Picsellia
  • Deploy and monitor a ModelVersion trained with Picsellia
  • Review Prediction done by the ModelVersion and enrich the training DatasetVersion
  • Retrain your ModelVersion with the enriched DatasetVersion

Once mastering the main core capabilities and depending on your needs, you can go further into the Picsellia features by relying on the extended platform documentation. (for instance, integrate your own model into Picsellia, apply processing on your Dataset etc..)

4. Vocabulary

Before deepening dive into the documentation, here is a bit of vocabulary used on the Picsellia platform:

  • Datalake Unique & shared place gathering all Data (images) related to an Organization
  • Data An image & associated Metadata contained in the Datalake
  • DataTag Additional Metadata that can be assigned to one or several Data in order to organize a Datalake
  • Dataset A placeholder for multiple Datasetversion
  • DatasetVersion A subset of Data inherited from the Datalake that will be annotated to be used later for a ModelVersion training
  • Asset An image & associated Metadata contained in a DatasetVersion, each asset is linked to its Data from the Datalake.
  • AssetTag Additional Metadata that can be assigned to one or several Asset in order to organize a DatasetVersion
  • Label An object that will store a class name and an id.
  • Annotation A set of Shapes annotated by one person.
  • Shape An annotated object can be a classification, a rectangle, a polygon, a line, or a point.

An extended glossary page is available here.