Does SNPE support converting recurrent neural network layers such as LSTM or GRU from Tensorflow models to DLC?
Recurent Neural Network Layers
Posted: Wed, 2017-10-25 07:34
Does SNPE support converting recurrent neural network layers such as LSTM or GRU from Tensorflow models to DLC?
Opinions expressed in the content posted here are the personal opinions of the original authors, and do not necessarily reflect those of Qualcomm Incorporated or its subsidiaries (“Qualcomm”). The content is provided for informational purposes only and is not meant to be an endorsement or representation by Qualcomm or any other party. This site may also provide links or references to non-Qualcomm sites and resources. Qualcomm makes no representations, warranties, or other commitments whatsoever about any non-Qualcomm sites or third-party resources that may be referenced, accessible from, or linked to this site.
Thanks for checking out SNPE. Unfortunately, this is not supported in the current SNPE version. While it is our policy not to discuss our roadmap in the forums, we are actively looking at providing LSTM support (including conversion from Tensorflow models).
Is it supported now on v1.8?
Yes. There is LSTM support in 1.8 on the GPU and CPU (no DSP runtime support yet, however).
Do you guys have any examples where an LSTM is used? Could you add a few examples to the SNPE SDK?
How to convert LSTM.pb file to .dlc file?
What should be the --input_dim ?
The LSTM model uses wav_data as input layer. Its trained with tensorflow and runs through tensorflow API on android.
does antone convert sucess?
i convert tensorflow model with lstm ,but my dlc file is 1KB. i dont know why. But if i set output node name before lstm ,it can convert sucess !!! SNPE version is 1.17.0
the lstm ops i use is :
rnn_cell_h18 = tf.contrib.rnn.BasicLSTMCell(num_units=1422, forget_bias=0.0)
Hi there! Is it possible to convert recurrent neural network layers (LSTM or GRU) from Tensorflow models to DLC? I'm using SNPE v1.27.1.
Thanks in advance.