No converter registered for op type: InstanceNormalization

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

CUDA:9.0
OS:centos
torch:1.1.0
tensorrt:4.0.1.6
model:pix2pixhd
cudnn:9.0
gpu:Tesla V100
https://github.com/NVIDIA/pix2pixHD
wen i run onnx on trt:
python test.py --name label2city_1024p --netG global --label_nc 0 --no_instance --onnx ./test.onnx
error:
input: “207”
input: “208”
input: “209”
output: “210”
op_type: “InstanceNormalization”
attribute {
name: “epsilon”
f: 1e-05
type: FLOAT
}
terminate called after throwing an instance of ‘std::out_of_range’
what(): No converter registered for op type: InstanceNormalization
Aborted

2

It looks like you are using ops not supported by TensorRT. Would recommend to update the TRT to latest version.
Please reference support matrix guide for a list of supported ops. Support Matrix :: NVIDIA Deep Learning TensorRT Documentation