Hello,
I'm using Intel FPGA AI 2023.2 on ubuntu 20.04 host computer and trying to infer a custom CNN in a Intel Arria 10 SoC FPGA.
Due to some accuracy drops, I replaced the ReLU activation layers with the PReLU activation layer in my mode, supported by the Intel FPGA AI Suite.
The IR model is generated, but the accuracy drops from 97.46% to 63.20%. The model optimizer function is as follows:
mo
--saved_model_dir "{path_savedModelPath}"
--input_shape "{lst_inputShape}"
--model_name "{str_modelName}"
--output_dir "{path_irTargetPath}"
--use_new_frontend
Please note that --use_new_frontend is used, but the IR model is not generated otherwise.
The architecture is A10_FP16_Generic.arch. The options used for dla_compiler are as follows:
dla_compiler
--march "{path_archPath}"
--network-file "{path_xmlPath}"
--o "{path_binPath}"
--foutput-format=open_vino_hetero
--fplugin "HETERO:FPGA,CPU"
--fanalyze-performance
--fdump-performance-report
--fanalyze-area
--fdump-area-report
But the graph is no compiled and the following error is displayed:
Layer (Name: StatefulPartitionedCall/model/p_re_lu_7/add_0_48_result, Type: Result) WRONG MESSAGE: Layer (Name: StatefulPartitionedCall/model/p_re_lu_7/add, Type: Eltwise) WRONG MESSAGE:
Layer (Name: StatefulPartitionedCall/model/p_re_lu_7/mul, Type: Eltwise) is not supported:
FPGA plugin: 'Constant' nodes are not supported as input of nodes Eltwise
Layer (Name: StatefulPartitionedCall/model/p_re_lu_7/Relu_1, Type: Relu) WRONG MESSAGE:
Layer (Name: StatefulPartitionedCall/model/p_re_lu_7/Neg_1, Type: PowerIE) WRONG MESSAGE:
Layer (Name: Constant_2809, Type: Constant) is not supported:
Error occurred.
../compiler/aot_plugin/src/dla_executable_network.cpp:134 Graph is not supported on FPGA plugin due to existance of layer (Name: StatefulPartitionedCall/model/p_re_lu_7/mul, Type: Eltwise)
in topology. Most likely you need to use heterogeneous plugin instead of FPGA plugin directly.
Find the model in the following link https://consigna.ugr.es/?s=download&token=11ed96b6-769d-4875-b8c7-0b73c1d0149a
Same problem still present in OpenVINO 2022.3.1 + Intel FPGA AI Suite 2023.3
