Enable openvino deployment

[sungchul2]

This PR includes:

• Enable validation with deployed models to onnx and openvino (trt and deploy are not tested.)

  $ python yolo/lazy.py \
      task=validation \
      name=AnyNameYouWant \ # AnyNameYouWant
      device=cpu \ # hardware cuda, cpu, mps
      model=v9-c \ # model version: v9-c, m, s
      task.fast_inference=onnx # onnx, openvino, trt, deploy

  • Save deployed models into weights/ instead of root

    weights
    ├── v9-c.bin
    ├── v9-c.onnx
    ├── v9-c.pt
    └── v9-c.xml

• Enable openvino deployment
  • To deploy the model to openvino

  $ pip install openvino
  $ python yolo/lazy.py task=validation model=v9-c device=cpu task.fast_inference=openvino
  ...
  [08/28 12:50:24] WARNING | __ Error loading OpenVINO model: Exception from src/inference/src/cpp/core.cpp:90:
  Check 'util::directory_exists(path) || util::file_exists(path)' failed at src/frontends/common/src/frontend.cpp:113:
  FrontEnd API failed with GeneralFailure
  ir: Could not open the file: "weights/v9-c.xml"
                                                                                                                                                                                                                                                                                                                          
  [08/28 12:50:26]   INFO  | 📥 OpenVINO model saved to weights/v9-c.xml
  ...
  
  # for fp16 compression
  $ python yolo/lazy.py task=validation model=v9-c device=cpu task.fast_inference=openvino task.precision=FP16
  ...
  [08/28 12:59:35] WARNING | __ Error loading OpenVINO model: Exception from src/inference/src/cpp/core.cpp:90:
  Check 'util::directory_exists(path) || util::file_exists(path)' failed at src/frontends/common/src/frontend.cpp:113:
  FrontEnd API failed with GeneralFailure:
  ir: Could not open the file: "weights/v9-c_fp16.xml"
  
  [08/28 12:59:37]   INFO  | 📥 OpenVINO model saved to weights/v9-c_fp16.xml
  ...

• Enable post-training quantization using nncf

  $ pip install nncf
  $ python yolo/lazy.py task=validation model=v9-c device=cpu task.fast_inference=openvino task.ptq=True # ptq to uint8
  ...
  [08/28 10:25:00]   INFO  | 🚀 Using OpenVINO as MODEL frameworks!
  INFO:nncf:NNCF initialized successfully. Supported frameworks detected: torch, onnx, openvino
  ...
  Statistics collection _____________________________________________________________________________________________________ 100% 300/300 _ 0:24:38 _ 0:00:00
  Applying Fast Bias correction _____________________________________________________________________________________________ 100% 291/291 _ 0:00:37 _ 0:00:00
  ...

TODO:

• Direct OpenVINO IR conversion without ONNX model