Assertion error while converting MobileNet V2 to TRT

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

Description

I am trying to follow GitHub - jkjung-avt/tensorrt_demos: TensorRT MODNet, YOLOv4, YOLOv3, SSD, MTCNN, and GoogLeNet to convert custom trained SSD Mobilenet V2 model to TensorRT format. I get this error while running the conversion.

#assertion/opt/tensorrt/TensorRT/plugin/nmsPlugin/nmsPlugin.cpp,246

I have tried changing the inputOrder on this line https://github.com/jkjung-avt/tensorrt_demos/blob/a061e44a82e1ca097f57e5a32f20daf5bebe7ade/ssd/build_engine.py#L58 to [0, 2, 1] as the pbtxt file generated contains the following format


id: "NMS"
    inputs: "Squeeze"
    inputs: "concat_priorbox"
    inputs: "concat_box_conf"

Environment

TensorRT Version: 7.0.0
Tensorflow. Version: 1.15
@AastaLLL
I am attaching my config file below as well.
pipeline.config (5.7 KB)

2

Hi,
UFF and Caffe Parser have been deprecated from TensorRT 7 onwards, hence request you to try ONNX parser.
Please check the below link for the same.

Thanks!

4

ONNX parser doesn’t support few layers and UFF parser have shown to be working on the current ssd_mobilenet_v2 graph.

I think this is a dimension mismatch error. Any help with this would be helpful

5

Hi,

Please refer official TensorRT object detection API models, which use ONNX parser and also support SSD MobileNet v2.

Thank you.

6

Thanks

But this doesn’t support Tensorrt 7.0 which I currently want to use. I need help with tensorrt 7

7

Sorry, TensorRT 7 is a very old version. UFF and Caffe parsers are deprecated from TRT7 onwards.
We recommend you use the latest TensorRT version.
https://developer.nvidia.com/nvidia-tensorrt-8x-download

Thank you.