@yinghai @benbarsdell @kevinch-nv
Hello, we now have a project based on InsightFace and want to accelerate with TensorRT. However, InsightFace is based on the Mxnet model, so it needs to be converted to the Onnx model. Among them, the model requires a Prelu activation function, but TensorRT-Onnx does not support Prelu operation. What I want to know is, when will the TensorRT-Onnx project support Prelu under the Onnx model, is there a development plan? Thank you!
@Yangqing @Maratyszcza @yinghai @nswamy @flyunix
Hello, we now have a project based on InsightFace and want to accelerate with TensorRT. However, InsightFace is based on the Mxnet model, so it needs to be converted to the Onnx model. Among them, the model requires a Prelu activation function, but TensorRT-Onnx does not support Prelu operation. What I want to know is, when will the TensorRT-Onnx project support Prelu under the Onnx model, is there a development plan? Thank you!
@flyunix onnx-tensorrt supports customized ops through plugins. Please take a look at https://github.com/onnx/onnx-tensorrt/blob/master/FancyActivation.cu for an example.
@yinghai no useful.
@yinghai no useful.
when will the TensorRT-Onnx project support Prelu under the Onnx model, is there a development plan? Thank you!
@flyunix PRELU is fully supported now in TensorRT 6.0. Please download the latest binaries and try importing your model. Closing this issue now, thanks!
@flyunix onnx-tensorrt supports customized ops through plugins. Please take a look at https://github.com/onnx/onnx-tensorrt/blob/master/FancyActivation.cu for an example.
The link is not reachable
Most helpful comment
@flyunix onnx-tensorrt supports customized ops through plugins. Please take a look at https://github.com/onnx/onnx-tensorrt/blob/master/FancyActivation.cu for an example.