Custom Layer Tensorflow support for TensorRT 3

Is TensorRT provides a Custom Layer C++ API for Tensorflow to inference UFF file on DrivePX2?

Dear krunal.vaghani,

Unfortunately, if use any network that requires custom layer in TensorRT, cannot use DriveWorks API for DNNs.



One mode thing, In TensorRT sample list, there is one samplePlugin example which shows that we can implement a custom caffe layer using Plugin API of TensorRT using NvCaffeParser.

So, you are saying that, if I use this custom layer DNN in TensorRT for example pedestrian detection network and that will not support on DrivePX2?

As for custom layered caffe networks, TensorRT has Plugin API so, for custom layered Tensorflow Model there is no way I mean like Plugin API that we can use TensorFlow model to inference on drivePX2?


so I would like to have a clear statement from you SteveNV. Is it possible to use a Tensorflow model with the UFF parser together with custom layers and do the integration on the DRIVE PX2? Yes or No? If yes, are there any examples? If no, is it planned in the near future to support plugins for Tensorflow models or do we have to switch to Caffe framework?


We are also wondering the same thing.

Dear All,

Could you please refer to the following link for your topic? Thanks.