Apertif Fast Radio Burst classification model

This is the Apertif classification model for dynamic spectra, also known as frequency-time data. It accepts input of shape [batch, channels, time, 1], where channels must be 32 and time must be 64. The input data is expected to be normalized per channel, e.g. by subtracting the median of the channel and dividing by its standard deviation. 

The model is in ONNX format, which can be run as-is with onnxruntime or converted to other formats for use with tools such as Tensorflow or Pytorch.