Extend AI Machine Learning Engineer Assessment

Congrats on making it this far! This test is to give you a small glimpse on the nature of the problems you could solve at Extend AI. Before diving in the complex nature of spacetime AI models (4D), we propose to start your journey at the 2D texture level.

We are pushing technology and we usually go after difficult problems to solve that usually requires between 2 weeks to 6 months of efforts. Having a good methodology to understand the problem, evaluate and test hypotheses, documenting and taking the solution to production is therefore essential to have fun working at Extend AI.

We like to think outside of the box with a first principles perspective.

We hope you'll enjoy solving this problem!

Challenge

• Given unlabeled images of wood slices (i.e. sections) in the data folder, create a ML model to identify and segment anomalies (i.e.knots, spots, etc).
• Present it as one or more Jupyter Notebooks.
• Describe your steps, we're really interested in how you tackle problems.
• Python and PyTorch preferred.
• Limit as much as possible the usage of classical CV. We are interested in seeing how you would approach this problem from a deep machine learning perspective.
• As you’ll see, the quantity of data is limited on purpose. Capturing the physical world is challenging and we want to see how creative you will be to solve this important and frequent data problem.
• Feel free to contact us for more information and clarifications if needed. You'll receive an invite to join a Slack channel so you can get in touch with us during the process.
• After you finish the challenge, propose how you would improve the model in the near future and how you would transfer learnings to other types of surfaces (i.e. not wood).

How to do the assessment

• Fork this repo.
• Code and document in your fork.
• Send a pull request once you’re done.