Openvino IE(Inference Engine) python samples - NCS2
before you start, make sure you have
- Dev machine with Intel 6th or above Core CPU (Ubuntu is preferred, a Win 10 should also work)
- Openvino 2018R5 or later installed and configured for NCS devices
- physical NCS2 VPU (the first gen NCS should also work, with a much lower perf)
- mobilenet-ssd caffe model downloaded, IR model file generated with MO ( what is IR/MO?)
how to run (remember to source /opt/intel/computer_vision_sdk/bin/setup_vars.sh in each new terminal)
- the basic "hello world" sample using IE sync API (you get a FPS around 16FPS on a i5-7200u in terminal outputs):
python3 sync_api.py
- the async sample using IE async API (this will boost you to 29FPS on a i5-7200u):
python3 async_api.py
- the 'async API' + 'multiple threads' implementation (this will boost you to 39FPS on a i5-7200u):
python3 async_api_multi-threads.py
- the 'async API' + 'multiple threads' + 'simutaneous inference requests' implementation (this will boost you to 46FPS on a i5-7200u if $requests_number is 4):
python3 async_api_multi-threads_multi-requests.py [requests_number]
- the 'async API' + 'multiple threads' + 'simutaneous inference requests' + 'multiple ncs' implementation (this will boost you to 87FPS on a i5-7200u if $ncs_number is 2 and $requests_number is 4), please note that we need to equip 2 or more NCS2 here
python3 async_api_multi-threads_multi-requests_multi-ncs.py [ncs_number] [requests_number]
- the 'async API' + 'multiple processes' + 'simutaneous inference requests' + 'multiple ncs' implementation (this will boost you to 89FPS on a i5-7200u if $ncs_number is 2 and $requests_number is 4), please note that we need to equip 2 or more NCS2 here
python3 async_api_multi-processes_multi-requests_multi-ncs.py [ncs_number] [requests_number]