Hansel Yang

> @kiritigowda @hansely the validation_app rali part was written by me. I can change it to mention rocal. It's only naming. the APIs I have already changed @LakshmiKumar23 Can you...

@kiritigowda Isn't caffe and onnx independent of each other? How does changing the onnx version affect caffe?

@kiritigowda Will update the issue after verification tomorrow

@kiritigowda Using protobuf==3.20.2 / onnx==1.13.0 version works well for both caffe & onnx models. We can close the issue after updating the setup script.

``` -- ******** Summary ******** -- CMake version : 3.20.4 -- CMake command : /usr/bin/cmake -- System : Linux -- C++ compiler : /usr/bin/c++ -- C++ compiler version : 7.5.0...

Hi @atamazov . You'll have to build [MIVisionX](https://github.com/GPUOpen-ProfessionalCompute-Libraries/MIVisionX) first. You can either build it on your system or use the [docker](https://github.com/GPUOpen-ProfessionalCompute-Libraries/MIVisionX#docker). After building MIVisionX, you can try running the unit...

Both slice & batchnorm call the same function but with different parameters. Slice pass nullptr for mean & varaince and 0 for epsilon.

Batchnorm: `miopenBatchNormalizationForwardInference(miopenHandle, miopenBNSpatial, &data->alpha, &data->beta, data->input_desc, data->input_mem, data->output_desc, data->output_mem, data->bnScaleBiasMeanVarDesc, data->bnScale, data->bnBias, data->bnMean, data->bnVariance, data->eps)` Slice: `miopenBatchNormalizationForwardInference(miopenHandle, miopenBNSpatial, &data->alpha, &data->beta, data->input_desc, data->input_mem, data->output_desc, data->output_mem, data->bnScaleBiasMeanVarDesc, data->bnScale, data->bnBias, nullptr, nullptr, 0));`...

@muralinr I understand. However, on [miopen documentation](https://rocmsoftwareplatform.github.io/MIOpen/doc/html/batchnorm.html?highlight=miopenbatchnorm#miopenbatchnormalizationforwardinference) it says passing nullptr for mean & variance is allowed. They will just not be used.

Thanks. Closing the issue.