My Environment
Hardware: NXP i.XM8MP EVK A01
Software: Android version 10
insightface pytorch project link：https://github.com/deepinsight/insightface/tree/master/recognition/arcface_torch
pytorch pre-train model download link：https://onedrive.live.com/?cid=4a83b6b633b029cc&id=4A83B6B633B029CC%215577&authkey=!AFZjr283nwZHqbA
I use benchmark run this model resnet_r100_quant_2.5.0.tflite then output message as follows.
./android_aarch64_benchmark_model_plus_flex --num_threads=4 --graph=resnet_r100_quant_2.5.0.tflite --use_nnapi=true --enable_op_profiling=true > resnet_r100_quant_2.5.0.txt
05-16 19:18:43.179 I//vendor/bin/hw/android.hardware.neuralnetworks@1.2-service-vsi-npu-server( 305): getSupportedOperations exit
05-16 19:18:43.181 I//vendor/bin/hw/android.hardware.neuralnetworks@1.2-service-vsi-npu-server( 305): getSupportedOperations
05-16 19:18:43.181 I//vendor/bin/hw/android.hardware.neuralnetworks@1.2-service-vsi-npu-server( 305): reject TRANSPOSE because >=4D or transposed on Batch not supported
05-16 19:18:43.181 I//vendor/bin/hw/android.hardware.neuralnetworks@1.2-service-vsi-npu-server( 305): reject PAD because pad value for 3rd or 0rd/3rd dimensions
05-16 19:18:43.181 I//vendor/bin/hw/android.hardware.neuralnetworks@1.2-service-vsi-npu-server( 305): reject TRANSPOSE because >=4D or transposed on Batch not supported
05-16 19:18:43.181 I//vendor/bin/hw/android.hardware.neuralnetworks@1.2-service-vsi-npu-server( 305):
05-16 19:18:43.181 I//vendor/bin/hw/android.hardware.neuralnetworks@1.2-service-vsi-npu-server( 305): getSupportedOperations exit
05-16 19:18:43.182 I//vendor/bin/hw/android.hardware.neuralnetworks@1.2-service-vsi-npu-server( 305): getSupportedOperations
05-16 19:18:43.183 I//vendor/bin/hw/android.hardware.neuralnetworks@1.2-service-vsi-npu-server( 305): reject TRANSPOSE because >=4D or transposed on Batch not supported
05-16 19:18:43.183 I//vendor/bin/hw/android.hardware.neuralnetworks@1.2-service-vsi-npu-server( 305): reject TRANSPOSE because >=4D or transposed on Batch not supported
This model is downloaded from the insightface pytorch project.
Step 1：use the pytorch tool to convert pytorch to onnx (pytorch 1.7.1)
Step 2：use onnx-tf to convert onnx to pb. (onnx-tf 1.6.0)
Step 3：use TensorFlow to convert pb to tflite (resnet_r100_quant_2.5.0.tflite). (TensorFlow 2.5.0)
How to solve the issue as follows
reject TRANSPOSE because >=4D or transposed on Batch not supported
reject PAD because pad value for 3rd or 0rd/3rd dimensions
Attach file is the result of benchmark.
