G-sensor_Example

• Gesture Recognition Magic Wand Training Scripts.
• The scripts in this directory can be utilized to train a TensorFlow model that classifies gestures using accelerometer data. The code is designed for TensorFlow 2.0. The resulting model has a size of less than 20KB. This project was inspired by the Gesture Recognition Magic Wand project by Jennifer Wang.

---

1. First step

1. Install virtual env

• If you haven't installed NuEdgeWise, please follow these steps to install Python virtual environment and choose NuEdgeWise_env.
• Skip if you have already done it.

2. Running

• The magic_wand_start.ipynb notebook will help you prepare data, train the model, and finally convert it to a TFLite and C++ file.

2. Work Flow

1. Dataset

• The dataset consists of accelerometer readings in three dimensions: x, y, and z, collected from various gestures.
• Users can collect their own data by running the code m467 sensor_collect on the m467 EVB.
• magic_wand_start.ipynb will assist you in collecting and preparing the data.

2. Training

• Use magic_wand_start.ipynb to train the model locally.

Training in Colab

• Utilize magic_wand_colab.ipynb to upload the dataset and train on Google Colab.

3. Test & Deployment

• Use magic_wand_start.ipynb to convert to TFLite and C++ files.

4. Vela Compiler (Optional)

• If device supports ARM NPU, like U55, please use vela\ to convert *quantized.tflite.

3. Inference code

• The ML_SampleCode repositories are private. Please contact Nuvoton to request access to these sample codes. Link
  • ML_M460_SampleCode (private repo)
  • tflu_magicwand: Real-time inference
  • tflu_magicwand_sensor_collect: Data collection firmware