[Python] QUALCOMM NEURAL NETWORK PROCESSING ENGINE (QNPE) NETWORK CREATION FAILED [closed]

I want to use snpe to inference unet model format dlc on qualcomm board, I convert model to onnx formart, then convert to dlc using snpe version v2.32.0.250228

My convert onnx code:

torch.onnx.export(
model,
dummy_input,
onnx_path,
export_params=True,
opset_version=opset_version,
do_constant_folding=True,
input_names=['input'],
output_names=['output'],
)


My convert dlc bash on window:

python "$SNPE_ROOT\bin\x86_64-windows-msvc\snpe-onnx-to-dlc" \
--input_network "model.onnx" \
--input_type input image \
--input_layout input NCHW \
--input_dim input "1, 3, 256, 256" \
--input_dtype input float32 \
--out_node output \
--output_path "model.dlc"


And I run on qualcomm board:

./snpe-net-run \
--container model.dlc \
--input_list input_list.txt \
--input_name input \
--input_dimensions "1,3,256,256" \
--use_dsp


I get that error:

Logging level is : SNPE_LOG_LEVEL_ERROR
/prj/qct/webtech_scratch20/mlg_user_admin/qaisw_source_repo/rel/2.32.0/release/snpe_src/avante-tools/prebuilt/dsp/hexagon-sdk-4.2.0/ipc/fastrpc/rpcmem/src/rpcmem_android.c:38:dummy call to rpcmem_init, rpcmem APIs will be used from libxdsprpc
error_code=401; error_message=Network creation has failed. inferOutputTensorShapes: Inconsistency in dynamic axis shapes. Cannot concat tensor with shapes: IrTensorShape(dims = [1,4294967294,32,512] , dy_axes=[]) and IrTensorShape(dims = [1,4294967295,32,512] , dy_axes=[]) on axis 3; error_component=Dl Network; line_no=67; thread_id=547649269776
/prj/qct/webtech_scratch20/mlg_user_admin/qaisw_source_repo/rel/2.32.0/release/snpe_src/avante-tools/prebuilt/dsp/hexagon-sdk-4.2.0/ipc/fastrpc/rpcmem/src/rpcmem_android.c:42:dummy call to rpcmem_deinit, rpcmem APIs will be used from libxdsprpc


Can someone help me for that?

Continue reading...