-3

I need some of the implementation of python in c++. Because of complexity in c++, I could not able to figure out how to achieve it in a simply and fastest way.

Note: I can implement in c++ using Eigen linear algebra library, because of time limit, I can not do this now.

So my question is, will it be possible to achieve the same implementation, especially, numpy part in the following link voxelization implementation using numpy in tensorflow c++ API or can I implement using tensorflow python and create a computational graph for the above implementation and link the output of per-processing graph to model graph as an input to use on hardware?

Use case of above question:

I have a neural network model for object detection using lidar point cloud. In the model author is doing the per-processing of data outside the graph. Since I need to provide the end model in C++, I thought of implementing the per-processing step in tensorflow C++ API. But after doing some research online, I got to know tensorflow c++ api does not support all the linear algebra operations yet like in tensorflow python API(one of my colleague told me this).

Any suggestions are most welcome.

Thanks in advance

Suri
  • 21
  • 5
  • your question is not clear. The link you have provided in your post does not even use tensorflow. Please edit your post and make your question clean. – CrazyBrazilian Jan 02 '21 at 05:14
  • @CrazyBrazilian question is, possibility of implementing the code in the link using tensorflow c++ API, instead of eigen library like numpy in python? I know it is bit of confusing. – Suri Jan 02 '21 at 09:04
  • i did look into the code ( not a very deep analyses ) and yes, you can easily write that code using tensorflow ( python or c++ ). Basically, you will transform that to a computational graph. – CrazyBrazilian Jan 02 '21 at 15:26
  • @CrazyBrazilian thank you so much for the hint. I can build the computational graph using python now. – Suri Jan 29 '21 at 13:53

1 Answers1

0

I found solution to my question. Unlike in python, there is no user friendly library, which can be used along with tensorflow C++ API. But I found some of the linear algebra libraries. This enter link description here provodes available libraries for c++. Out of all, xtensor is friendly to use like numpy, still lacks some of the features.

While choosing any library, one need to know how stable the libraries is. Especially for production code.

Suri
  • 21
  • 5