Do we have to use BatchedNMSDynamic_TRT in the model output by ONNX?

[mito-w]

I am trying to transfer train a yolo_v4_tiny model using TAOtoolkit and convert the exported model in ONNX format using VectorBlox-SDK.
In doing so, I get an error regarding BatchedNMSDynamic_TRT and cannot proceed.

OpenVINO does not support the following ONNX operations: BatchedNMSDynamic_TRT

Do I have to use this BatchedNMSDynamic_TRT operator when using TAOtoolkit for transition learning?
Or is there another solution?

[Morganh]

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

@mito-w

Do I have to use this BatchedNMSDynamic_TRT operator when using TAOtoolkit for transition learning?

When use TAO for transfer learning, as shown in the notebook or user guide, users can set .hdf5 file or .tlt file(encrypted hdf5 file) as the pretrained model. It is not related to the .onnx file.

@mito-w

In doing so, I get an error regarding BatchedNMSDynamic_TRT and cannot proceed.

OpenVINO does not support the following ONNX operations: BatchedNMSDynamic_TRT

You can trim the exported onnx file to a new onnx file. Then perform NMS by yourself. Refer to Errors while reading ONNX file produced by TAO 5 - #10 by Morganh.