A typical AI model development process consists of ‘data preparation’, ‘model training’, and ‘model evaluation and validation’. You need to repeat this process until you meet the criteria and this repetitive process can take up to several months or even a year. Moreover, developing a production-level AI model to deploy on a particular device requires additional time. The model must meet the performance requirements on the target hardware (latency, memory footprint, etc.), and building this kind of model becomes more challenging if pre-processing, post-processing, and other SW should be included.
NetsPresso automates the model development process so that engineers can easily obtain an optimized model with desired performance. In particular, we are concentrating on the edge AI area where NetsPresso's technology is most needed.
NetsPresso consists of three modules in one pipeline. There are Model Searcher that automatically creates a model according to the target device, Model Compressor that improves the calculation complexity and inference speed of the model with a lightweight technique, and Model Launcher that converts and packages the model so that it can be deployed directly to the target device.
If you need production-level models, we also provide our professional service.
Please contact [email protected].
Updated 8 months ago