Description#
During my time as working student at Fraunhofer IPA, I’ve developed a denture/tooth classifier as part of a feasibility study for a research project. The classifier follows dental notation and is trained on a handcrafted dataset of dental images in various poses.
The result was a CNN with an accuracy of over 98% and the decision to continue the project to market maturity with an even wider array of dental variants. Due to my graduation and subsequent employment at Hochschule der Medien Stuttgart, I was not able to continue working on this project.
Technologies#
- EfficientDet: The backbone object detection model used for the classifier.
- PyTorch: Implementation of the classifier and training pipeline.
- Pydantic: Used for settings management.
- Basler: The camera used for capturing the training images.
