A roadmap for Scivision

Metadata

Editors:

quantumjot, scotthosking, sebastianahnert (project PIs), ots22 (who created this draft)

Status (raw | draft | stable | deprecated | retired):

draft

Description

This document summarises the main priorities of the Scivision team based at the Alan Turing Institute, for the current phase of the project supporting Scivision development, that started in January 2023.

Scivision originated at the Alan Turing Institute, and development is actively hosted there, but the project welcomes participation from anyone in the wider community. Any contributions related to Scivision’s mission are gladly received, whether or not they relate to the items noted below.

This roadmap is to be understood as listing areas that are likely to receive attention from the Turing team, and is provided in order to help users and contributors understand the direction of the project, given the Turing team’s stake in it.

Importantly, the roadmap doesn’t commit anyone to any particular deliverables, there aren’t specific dates associated with the items listed here, and it may be modified (it is likely to remain a perpetual ‘draft’).

Priority areas

Core features

• Data loaders (potentially replacing or supplementing Intake)

• Catalog schema improvements and model/datasource metadata; catalog versioning

• Better specified requirements for models and datasources

• More robust handling of model dependencies

Community engagement

• A community calendar

• Running events, including a regular community call

• Making Scivision a hub for best practices in sharing reproducible CV models

• Informational videos (overview, getting started, task-focussed tutorials)

• Informal support to researchers wishing to share their models and datasets

Web interface

• Model/dataset matchmaking, graphical representation of models and dataset interoperability

• Improved ‘model cards’

  • For example, show extra installation instructions when these are required, and give an example of using the model in Python

• Improvements to the ‘new model/new datasource’ interface

Automation

• Template repositories for computer vision models or datasources that are ready for inclusion in the catalog (‘cookie cutters’).

• Additional consistency and integrity checks, including

  • URLs in the catalogs

  • scivision gallery notebooks (from current scivision projects)

  • checked annotations (decorators)

Supporting Scivision use-cases

• Support to researchers at the Turing and collaborators

• Growing the catalog

  • Helping users to share their models and data through the catalog

  • Targetting models for the following tasks in particular (although not exclusively):

    • object tracking

    • super-resolution

    • shape analysis