This is a project regarding edge implementation for human activity recognition (HAR).
- Built offline models (CNN and LSTM) is in 'Offline_models' folder.
- The convertion and interpreter codes can be directly seen in the main interfacne.
- Two quantised models using differnet quantisaion strategies (full integer with float fall back, and full integer only) are generated, namely 'tflite_model_quant_fullint', and 'tflite_model_quant_uint8' (which can be seen in 'TFlite_models' folder).
- The code of the deployment of these two models are in 'Android_HAR_app' folder.
- 'e4link_har' folder contains the code of the final real-time CNN model calssifier application.