This repo exists as an exercise to prepare a pretrained model (superpoint) to be used on different hardware platforms. Specifically, we convert superpoint's pytorch implementation to onnx, intel targets (using openVINO), and finally the movidius inference accelerator. Future work may be conversion for Coral inference accelerator. To use openVINO, we have re-implemented parts of superpoint in numpy for layers and other parts of execution that are not available out-of-the-box.
To get onnx representation, we have a script to produce the onnx version of superpoint.
# Usage
python3 convert_onnx.py <path_to_superpoint_weights_file> <height_of_input_image> <width_of_input_image>
# Example
python3 convert_onnx.py weights/superpoint_v1.pth 100 100
# Output will default to output/ directory.
We use openVINO tools to build. This is done expecting the openVINO toolkit is already installed. If you don't have it, go here If you want to build to movidius, use openVINO 2020.1 or 2020.2 and follow the directions below. Example of usage can be seen in run_openvino.py
# Usage
python3 /opt/intel/openvino/deployment_tools/model_optimizer/mo.py --input_shape [1,1,<height_of_input_image>,<width_of_input_image>] --input_model <path_to_onnx_file> --output_dir output --data_type FP16
# Example
python3 /opt/intel/openvino/deployment_tools/model_optimizer/mo.py --input_shape [1,1,100,100] --input_model output/superpoint_100x100.onnx --output_dir output --data_type FP16
# Output .bin and .xml files will be in output/
We generate this representation using openVINO, and use it using depthai, which can be installed here. Example of usge can be seen in run_oak.py. It expects to be running a MegaAI/OAK-1 module.
# Usage
/opt/intel/openvino/deployment_tools/inference_engine/lib/intel64/myriad_compile -m <path_to_xml_file> -o <path_to_blob_file> -ip U8 -VPU_MYRIAD_PLATFORM VPU_MYRIAD_2480 -VPU_NUMBER_OF_SHAVES 4 -VPU_NUMBER_OF_CMX_SLICES 4
# Example
/opt/intel/openvino/deployment_tools/inference_engine/lib/intel64/myriad_compile -m output/superpoint_100x100.xml -o output/superpoint_100x100.blob -ip U8 -VPU_MYRIAD_PLATFORM VPU_MYRIAD_2480 -VPU_NUMBER_OF_SHAVES 4 -VPU_NUMBER_OF_CMX_SLICES 4