Synonyms : Test Project – “Calibration batches” (cf Scale) – Training project (cf Kili) – Prototype
A POC project represents the first project that seeks to validate many assumptions before launching a production project. This project allows to adjust many variables such as:
- The quality standards expected by the client. These can be set by KPIs.
- The time spent per asset to be annotated (proportional to the expected quality, but inversely proportional to the annotator’s expertise)
- The balance between time and quality on annotations: it is always possible to improve the quality of annotations, however, the time per asset may grow accordingly.
- The type of annotation to be performed (bounding boxes, polygons, segmentation, etc.)
- The selection of an appropriate and well-adjusted annotation tool
- The selection and training of annotators and project managers adapted to the task.
- The definition of an annotation process (reviewing, consensus, etc.), necessary if the needed quality is very high.
Going through a POC Project makes it easier to scale: it limits errors and waste of time and money when scaling up.
At People for AI, we prefer to start with a paid POC project, in order to have sufficient budget to perform in-depth tests and studies of the annotation project.
The POC Project also allows to clarify the annotation guide and make it as exhaustive as possible. It also allows to properly train annotators from this guide.