Hi
I converted my Caffe model to IR using the latest version (2019 R1). Now I'm trying to use the calibration tool sample to convert my FP32 model to INT8. I'm able to load the model as FP32 and get the accuracy but it throws an error while loading the model for the first iteration of INT8, precisely in this line:
ExecutableNetwork executable_network = _pluginI8C.LoadNetwork(network, { { CONFIG_KEY(EXCLUSIVE_ASYNC_REQUESTS), CONFIG_VALUE(YES) } });
The message on the console:
[ INFO ] InferenceEngine: API version ............ 1.6 Build .................. 22443 [ INFO ] Parsing input parameters [ INFO ] Loading plugin API version ............ 1.6 Build .................. 22443 Description ....... MKLDNNPlugin [ INFO ] Loading network files [ INFO ] Preparing input blobs [ INFO ] Batch size is 1 [ INFO ] Collecting accuracy metric in FP32 mode to get a baseline, collecting activation statistics Progress: [....................] 100.00% done FP32 Accuracy: 95.45% [ INFO ] Verification of network accuracy if all possible layers converted to INT8 Validate int8 accuracy, threshold for activation statistics = 100.00 [ ERROR ] Inference problem: Supported primitive descriptors list is empty for node: Mul_722/FusedScaleShift_ ..\src\mkldnn_plugin\mkldnn_node.cpp:224 f:\intelswtools\openvino_2019.1.087\deployment_tools\inference_engine\include\details\ie_exception_conversion.hpp:71
Please let me know how to resolve this issue.
Thanks,
Indrajit
連結已複製
Hi Indrajit,
1. Firstly, can you rerun calibration tool with additional command line arguments (CPU extension), please:
-l C:\Users\sdramani\Documents\Intel\OpenVINO\inference_engine_samples_build_2017\intel64\Release\cpu_extension.dll
I hope everything should work with that.
2. Secondly, I would like to suggest to use Python Calibration Tool: https://docs.openvinotoolkit.org/latest/_inference_engine_tools_calibration_tool_README.html. Python Calibration Tool supports more datasets and topologies.
Best regards, Edward
Dear Indrajit, I already tried what Edward suggested (adding -l cpu_extension.dll). Still got the same error. Does it work for you ? If adding -l fails, it's a bug (which I've already filed).
Thanks,
Shubha
