The aim of the project is to provide an end-to-end solution for on-device training, inference and data collection for activity recongition based on TFlite Transfer Learning Pipeline. The corresponding blog post is available here.

• Python 3.5+
• Tensorflow 2.0.0rc0
• Numpy
• Pillow
• Scipy
• Android Studio

• Add support for pairing the app with a smartwatch and fine-tuning the model for a wearable device.
• Port SoundNet, add functionality for audio recording and tflite model conversion for handling dynamic size input.

If you are interested in contributing to this project, please submit a pull request or reach out at: aqibsaeed@protonmail.com.

The Heterogeneity Activity Recognition dataset is used for model pretraining. If you use this in your research, please cite their work and check the license.

If you find this project usefuly, please cite it as:

@misc{saeed2020recognition, 
  author = {Saeed, Aaqib},
  title = {On-device Learning of Activity Recognition Networks},
  year = {2020},
  journal = {aqibsaeed.github.io},
  url = {\url{https://gitHub.com/aqibsaeed/on-device-activity-recognition}}
}