Is compatible with hdf5 data layer?


I am training a non image data set in caffe and trying to convert it to corresponding cmsisnn calls using the nn_quantizer script. The input layer used in caffe is HDF5.

Post training the model in caffe and generating the caffemodel file, I run nn_quantizer on the model to generate cmsisnn API's. I face a problem that the script gets stuck in processing the hdf5 layer because apparently the script does not support hdf5 input layer.


allowed_layer_types = ['data','convolution','innerproduct','pooling','lrn','relu',\

When I try to convert any image recognition model using lmdb input format, the script works fine.

Is my understanding of the issue correct? If its correct then is there any alternate way for me to convert the non image data set model to cmsisnn calls?

Thanks for your time and suggestions.

- Yunas 

More questions in this forum