Skip to content

Latest commit

 

History

History
 
 

examples

README.md

What we have

The list below are the remiders for each example. For details, please see the docs under the folder individually.

  • auto_test is a PC based example, which is use by NNoM for travis CI. However, it is a very good demo for NNoM without borthering MCU. You can run it directly on your PC.
  • keyword_spotting is a regular convolution model to spot speech commands train by google speech dataset. This example use MFCC to extract the voice features, then use neural network to classified these commands into 30+ classes. The model has achieved around 90% Top 1 accuracy.
  • mnist-simple is an interactive example for RT-Thread using its Mesh shell. This example come with 10 embedded images allows users to do prediction through terminal. Chinese guides available.
  • octave-conv is to show how to construct the latest Octave convolution in Keras then deploy to NNoM.
  • uci-inception is an example using data from motion sensors and Inception structure. It is an interactive example using shell and Y-modem to transmit testing data.
  • rnn-denoise is a fixed point implementation of RNN speech noise suppression using the methodology provided by RNNoise. It implement an RNN network like RNNoise. It can run on PC like auto-test example.
  • uci-har-RNN is an example using data from motion sensors and Inception structure. However, use a stack of RNN layers to test the data. It can run on PC like auto-test example.
  • mnist-cnn is an entry level example using jupyter notebook following this Keras tutorial
  • mnist-densenet is a example showing how to use DenseNet with NNoM. This example can be compiled in PC using scons. It is also been use for Travis CI. Please check the Travis logging as well.

Recommendations

If you want to try it first on PC, start with auto_test and follow the guide in it. uci-har-rnn also provides c code which can run on PC.

If you are completely void in ML or Neural network, start with mnist-cnn and the external tutorial for Keras.

If you are trying to handle time sequence data (e.g. sensor measurement or voice), please check keyword_spotting and uci-inception.

If you are using RT-Thread, the very first example you should try is mnist-simple.

Environment

Recommended to use Tensorflow 2.0+ or Tensorflow 1.14+.

keyword_spotting requires a microphone on your development board and the driver of the mic, if you are not using 32L476GDISCOVERY