How do we work?
Our approach to delivering you with a high quality dataset can be broken down into a simple three stage process. This process aims to generate a dataset that resolves the gap between what the available real image dataset represents and what the deep learning model requires, to precisely and reliably solve the vision task at hand.
We use a ‘structured domain randomization’ technique to create a multitude of relevant images with high fidelity.
We integrate project and use case specific parameters and variations, experience based knowledge from customer teams, our domain know-how and the unending possibilities that our image generation technique offers, in our project process.
Unlimited possibilities in a
lean process
03 Steps
1
01. Scope and specification
In this phase we collect the relevant information related to the problem statement, scope of work, boundary conditions and success criteria.
Time / effort requirement : 1-2 meetings with the project in-charge and domain specialist (60 minutes).
2
02. Review and refine
Our experIn this phase, we set up the data pipeline and provide an initial dataset for review and approval. Change requests are handled in this stage.
Time / effort requirement : 1-2 meetings with the domain specialist and data team (60-90 minutes).
3
03. Dataset delivery
Voila! We generate the required dataset, that is ready for you to train or validate your deep learning model. Offboarding, advice on training with synthetic image data and review of success criteria is a part of this stage.
Time / effort requirement : 1-2 meetings with project in-charge and data team (30-60 minutes).
Outlook:
With a view to make our technology easily accessible to customers and to address privacy requirements, we are working towards offering a self service, subscription based image generation tool in the near future.
To know more regarding our current offering and for questions regarding the self service tool, please contact us.
Resources
Get a copy of brochure:
SI in a Nutshell
