CodeCamp #102: support SOLO deployment with OpenVINO

[DDGRCF]

Motivation

支持SOLO mmdet3.x Openvino导出

Modification

• support SOLO mmdet3.x Openvino export

Use cases (Optional)

python tools/deploy.py configs/mmdet/instance-seg/instance-seg_openvino_dynamic-800x1344.py \ 
                                    ${MMDETECTION_DIR}/configs/solo/solo_r50_fpn_1x_coco.py  \
                                    ${MMDETECTION_DIR}/demo/demo.jpg \
                                    --work-dir work_dirs/solo/instance-seg_openvino_dynamic-800x1344 \
                                    --device cpu

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit tests to ensure the correctness.
3. If the modification has a dependency on downstream projects of a newer version, this PR should be tested with all supported versions of downstream projects.
4. The documentation has been modified accordingly, like docstring or example tutorials.

[mm-assistant]

We recommend using English or English & Chinese for pull requests so that we could have broader discussion.

[hanrui1sensetime]

Please add unit test for the rewriter, test the performance using tools/test.py, and update the docs.

[hanrui1sensetime]

The ut in the CI is failed, may fix it.

[CLAassistant]

All committers have signed the CLA.

[grimoire]

Hi, We have refactor the rewriter. ctx has been move out of the arguments. You can still use ctx with FUNCTION_REWRITER.get_context() in the rewrite function.
And please sign the cla.

[grimoire]

LGTM.

[AllentDan]

In my testing, it failed at onnx2openvino stage. The error log goes like this:

subprocess.CalledProcessError: Command 'mo --input_model="work_dirs/openvino/det/solo/end2end.onnx" --output_dir="work_dirs/openvino/det/solo" --output="dets,labels,masks" --input="input" --input_shape="[1, 3, 800, 1344]" ' returned non-zero exit status 1.

The pipeline was okay for MRCNN FYI.

[DDGRCF]

In my testing, it failed at onnx2openvino stage. The error log goes like this:

subprocess.CalledProcessError: Command 'mo --input_model="work_dirs/openvino/det/solo/end2end.onnx" --output_dir="work_dirs/openvino/det/solo" --output="dets,labels,masks" --input="input" --input_shape="[1, 3, 800, 1344]" ' returned non-zero exit status 1.

The pipeline was okay for MRCNN FYI.

Strangely! I success in my server. And here is my export information.

12/17 19:54:44 - mmengine - INFO - Args for Model Optimizer: mo --input_model="work_dirs/solo/instance-seg_openvino_dynamic-800x1344/end2end.onnx" --output_dir="work_dirs/solo/instance-seg_openvino_dynamic-800x1344" --output="dets,labels,masks" --input="input" --input_shape="[1, 3, 800, 1344]" 
12/17 19:54:48 - mmengine - INFO - Model Optimizer arguments:
Common parameters:
        - Path to the Input Model:      mmdeploy/work_dirs/solo/instance-seg_openvino_dynamic-800x1344/end2end.onnx
        - Path for generated IR:         mmdeploy/work_dirs/solo/instance-seg_openvino_dynamic-800x1344
        - IR output name:       end2end
        - Log level:    ERROR
        - Batch:        Not specified, inherited from the model
        - Input layers:         input
        - Output layers:        dets,labels,masks
        - Input shapes:         [1, 3, 800, 1344]
        - Source layout:        Not specified
        - Target layout:        Not specified
        - Layout:       Not specified
        - Mean values:  Not specified
        - Scale values:         Not specified
        - Scale factor:         Not specified
        - Precision of IR:      FP32
        - Enable fusing:        True
        - User transformations:         Not specified
        - Reverse input channels:       False
        - Enable IR generation for fixed input shape:   False
        - Use the transformations config file:  None
Advanced parameters:
        - Force the usage of legacy Frontend of Model Optimizer for model conversion into IR:   False
        - Force the usage of new Frontend of Model Optimizer for model conversion into IR:      False
OpenVINO runtime found in:      openvino_2022.2.0.7713/python/python3.7/openvino
OpenVINO runtime version:       2022.2.0-7713-af16ea1d79a-releases/2022/2
Model Optimizer version:        2022.2.0-7713-af16ea1d79a-releases/2022/2
[ SUCCESS ] Generated IR version 11 model.
[ SUCCESS ] XML file: mmdeploy/work_dirs/solo/instance-seg_openvino_dynamic-800x1344/end2end.xml
[ SUCCESS ] BIN file: mmdeploy/work_dirs/solo/instance-seg_openvino_dynamic-800x1344/end2end.bin
[ SUCCESS ] Total execution time: 2.25 seconds. 
[ SUCCESS ] Memory consumed: 364 MB.

I think 800x1344 may be too large for some computer. Can you test in 300x300 openvino deploy config(I don't have other computer).

[AllentDan]

Tested OK in another machine. May update tests/regression/mmdet.yml

[AllentDan]

LGTM