When converting the yolov8-det model to ONNX using the NMS operator, the following error occurs when using trtexec to convert it into an engine model.

[alanxinn]

When converting the yolov8-det model to ONNX using the NMS operator, the following error occurs when using trtexec to convert it into an engine model.

CUDA 11.8
CUDNN 8.9.3
TensorRT 8.6.1.6

ultralytics 8.3.79

onnx 1.17.0 opset 19

error

[03/14/2025-11:03:27] [E] [TRT] [gatherNode.cpp::nvinfer1::builder::`anonymous-namespace'::computeGatherOutputExtents::76] Error Code 4: Internal Error (/Gather_2: rank of data tensor must be greater than or equal to 1)
[03/14/2025-11:03:27] [E] [TRT] ModelImporter.cpp:771: While parsing node number 248 [Gather -> "/Gather_2_output_0"]:
[03/14/2025-11:03:27] [E] [TRT] ModelImporter.cpp:772: --- Begin node ---
[03/14/2025-11:03:27] [E] [TRT] ModelImporter.cpp:773: input: "/ReduceMax_output_0"
input: "/model/model.22/Constant_output_0"
output: "/Gather_2_output_0"
name: "/Gather_2"
op_type: "Gather"
attribute {
name: "axis"
i: 0
type: INT
}

[03/14/2025-11:03:27] [E] [TRT] ModelImporter.cpp:774: --- End node ---
[03/14/2025-11:03:27] [E] [TRT] ModelImporter.cpp:777: ERROR: ModelImporter.cpp:195 In function parseGraph:
[6] Invalid Node - /Gather_2
[gatherNode.cpp::nvinfer1::builder::anonymous-namespace'::computeGatherOutputExtents::76] Error Code 4: Internal Error (/Gather_2: rank of data tensor must be greater than or equal to 1) load error: Invalid Node - /Gather_2 [gatherNode.cpp::nvinfer1::builder::anonymous-namespace'::computeGatherOutputExtents::76] Error Code 4: Internal Error (/Gather_2: rank of data tensor must be greater than or equal to 1)
tensorRT load mask onnx model successfully!!!...
[03/14/2025-11:03:27] [I] [TRT] BuilderFlag::kTF32 is set but hardware does not support TF32. Disabling TF32.
[03/14/2025-11:03:27] [E] [TRT] 4: [network.cpp::nvinfer1::Network::validate::2882] Error Code 4: Internal Error (Network must have at least one output)
try to save engine file now~~~

[lix19937]

rank of data tensor must be greater than or equal to 1

where 0 < num_elementwise_dims <= 1

see https://docs.nvidia.com/deeplearning/tensorrt/operators/docs/Gather.html#shape-information