Facts
- AI algorithm using transfer learning on a resnet18 pre-trained model
- 72 layers / 18 deep layers
- Learning data collected and edited directly in Forbeam
- Low data requirements thanks to unique augmentation algorithm
- Learning service running in the cloud:
- No expensive hardware required
- Latest version of model always available
- High level of privacy
- No storage of customer datasets and models
- Installation on premise possible in the futur
- Asynchronous model learning
- Detached from Forbeam loaded project
- Multiple learning jobs in parallel
- Subsequent retrieval of learned models
- Inference performed by the backend directly in memory
- No specific hardware required
- Performances depending on CPU
- Available on all platforms (Linux, Windows)