Convert a mask rcnn with tf's fronzen graph format to uff fail

I download below mask rcnn pretrained models from

  • mask_rcnn_resnet101_atrous_coco
  • mask_rcnn_resnet50_atrous_coco
  • mask_rcnn_inception_v2_coco

I extract the .gz file and use to convert .pb file to uff format just like below:
python ~/mrcnn-res101/frozen_inference_graph.pb

but convertion is failed,and python report below error info:
“uff.model.exceptions.UffException: Transpose permutation has op Sub, expected Const. Only constant permuations are supported in UFF.”

What’s wrong happened? How to solve it?

information of my system:
Ubuntu Linux 18.04
Cuda v10.0
CuDNN v7.5.0
TensorRT v5.0.2.6
NVIDIA Driver v415.27

Who can help me?


per engineering, there are several control flow ops and dynamic transposes, which are currently unsupported in TRT. Recommend trying TF-TRT instead.

Hi, Any resolution or work around for the above issue?