There is a newer version of the record available.

Published September 25, 2022 | Version 1.1.31
Software Open

Tools to convert ONNX files (NCHW) to TensorFlow format (NHWC)

Authors/Creators

Description

  • Support for --replace_erf_to_pseudo_erf option
  • Replace Erf not supported by TFLite with a primitive approximation
    • approximate expression
      eps = 1e-11
      x_abs = tf.math.abs(input_tensor)+eps
      sign = tf.math.divide(input_tensor, x_abs)
      a1 =  0.254829592
      a2 = -0.284496736
      a3 =  1.421413741
      a4 = -1.453152027
      a5 =  1.061405429
      p  =  0.3275911
      t = 1.0/(1.0 + p*x_abs)
      y = 1.0 - (((((a5*t + a4)*t) + a3)*t + a2)*t + a1)*t*tf.math.exp(-x_abs*x_abs)
      erf_tensor = sign*y
      tf_layers_dict[graph_node_output.name]['tf_node'] = erf_tensor
      

Full Changelog: https://github.com/PINTO0309/onnx2tf/compare/1.1.30...1.1.31

Notes

If you use onnx2tf in your research, please cite it using these metadata.

Files

PINTO0309/onnx2tf-1.1.31.zip

Files (378.0 kB)

Name Size Download all
md5:276d906b0600dc73b11529de00735e5d
378.0 kB Preview Download

Additional details

Related works