Can any provide a link to a mask rcnn model which will create valid engine file for tensort 6 or 7

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

Description

A clear and concise description of the bug or issue.
none of the mark rcnn model we have used can be run in tensorrt 6 or 7

Environment

nvidia jetson jp 4.3 or jp 4.4

TensorRT Version: 6 and 7
GPU Type: 410
Nvidia Driver Version: 10.2
CUDA Version:
CUDNN Version:
Operating System + Version:
Python Version (if applicable):
TensorFlow Version (if applicable):
PyTorch Version (if applicable):
Baremetal or Container (if container which image + tag):

Relevant Files

every conversion we have tryed has ended up with unsupported layers

Steps To Reproduce

try to run any mask rcnn against tensorrt 6 or 7 and get unsupported layers

Please include:

  • Exact steps/commands to build your repro
  • Exact steps/commands to run your repro
  • Full traceback of errors encountered
2

Sample’s model is based on the Keras implementation of Mask R-CNN and its training framework can be found in below link:

Please refer to steps mentioned in below link:
https://github.com/NVIDIA/TensorRT/tree/master/samples/opensource/sampleUffMaskRCNN#prerequisites

Known issues:
https://github.com/NVIDIA/TensorRT/tree/master/samples/opensource/sampleUffMaskRCNN#known-issues

Thanks

3

Hi,

Thanks for replying to my questions,
We have tried and implemented both the matterport mask rcnn repo and the sampleUffMaskRCNN tensorrt in both python environments and in docker containers and get the same errrors.

we have tried the suggested fix of cuda 10.1 in both python environment and in various nvidia docker containers.

we get an error on conversion

sampleUffMaskRCNN - No conversion function registered for layer: PyramidROIAlign_TRT

it is the same error in both docker and on bare metal

what information could be useful to you from us.
we are keen to learn and will supply anything you need from us to resolve this issue

thanks in advanced

4

Hi,

To generate the UFF model required for this sample, use a container/setup built with CUDA_VERSION=10.0 .

Thanks