Development and Evaluation of a Flutter Plugin for the Performant Execution of Deep Neural Networks in the ONNX Format on Edge Devices with NPUs

Context

The aim of the master thesis is the development of a Flutter plugin that enables the execution of Deep Neural Networks (DNNs) in ONNX format on mobile devices and is optimized for the use of Neural Processing Units (NPUs).

Objectives

Analysis of the requirements for the integration of ONNX models in Flutter applications Evaluation of existing frameworks for ONNX execution on mobile platforms (e.g. Android Neural Networks API, Core ML, etc.) Development of a Flutter plugin (possibly with platform-specific code in Kotlin/Swift/C++) for the execution of DNNs on NPUs Performance comparison: NPU vs. CPU/GPU execution Documentation and integration into a sample application

Prerequisites

A degree in computer science, software engineering, electrical engineering or a comparable subject Knowledge of Flutter/Dart as well as C++ or Java/Kotlin or Swift Interest in machine learning, mobile platforms and embedded AI Initial experience with ONNX, TensorFlow Lite or similar frameworks is an advantage Independent and structured way of working as well as the ability to work in a team