sod4onnx

Simple model Output OP Deletion tools for ONNX.

https://github.com/PINTO0309/simple-onnx-processing-tools

1. Setup

1-1. HostPC

### option
$ echo export PATH="~/.local/bin:$PATH" >> ~/.bashrc \
&& source ~/.bashrc

### run
$ pip install -U onnx \
&& python3 -m pip install -U onnx_graphsurgeon --index-url https://pypi.ngc.nvidia.com \
&& pip install -U sod4onnx

1-2. Docker

https://github.com/PINTO0309/simple-onnx-processing-tools#docker

2. CLI Usage

$ sod4onnx -h

usage:
    sod4onnx [-h]
    -if INPUT_ONNX_FILE_PATH
    -on OUTPUT_OP_NAMES [OUTPUT_OP_NAMES ...]
    -of OUTPUT_ONNX_FILE_PATH
    [-n]

optional arguments:
  -h, --help
        show this help message and exit.

  -if INPUT_ONNX_FILE_PATH, --input_onnx_file_path INPUT_ONNX_FILE_PATH
        Input onnx file path.

  -on OUTPUT_OP_NAMES [OUTPUT_OP_NAMES ...], --output_op_names OUTPUT_OP_NAMES [OUTPUT_OP_NAMES ...]
        Output name to be deleted to the models output OP.
        e.g.
        --output_op_names "output1" "output3"

  -of OUTPUT_ONNX_FILE_PATH, --output_onnx_file_path OUTPUT_ONNX_FILE_PATH
        Output onnx file path.

  -n, --non_verbose
        Do not show all information logs. Only error logs are displayed.

3. In-script Usage

>>> from sod4onnx import outputs_delete
>>> help(outputs_delete)

Help on function outputs_delete in module sod4onnx.onnx_model_output_deleter:

outputs_delete(
    input_onnx_file_path: Union[str, NoneType] = '',
    onnx_graph: Union[onnx.onnx_ml_pb2.ModelProto, NoneType] = None,
    output_op_names: Union[List[str], NoneType] = [],
    output_onnx_file_path: Union[str, NoneType] = '',
    non_verbose: Union[bool, NoneType] = False
) -> onnx.onnx_ml_pb2.ModelProto

    Parameters
    ----------
    input_onnx_file_path: Optional[str]
        Input onnx file path.
        Either input_onnx_file_path or onnx_graph must be specified.
        Default: ''

    onnx_graph: Optional[onnx.ModelProto]
        onnx.ModelProto.
        Either input_onnx_file_path or onnx_graph must be specified.
        onnx_graph If specified, ignore input_onnx_file_path and process onnx_graph.

    output_op_names: List[str]
        Output name to be deleted to the models output OP.
        e.g.
        output_op_names = ["output1", "output3"]

    output_onnx_file_path: Optional[str]
        Output onnx file path. If not specified, no ONNX file is output.
        Default: ''

    non_verbose: Optional[bool]
        Do not show all information logs. Only error logs are displayed.
        Default: False

    Returns
    -------
    outputops_deleted_graph: onnx.ModelProto
        onnx.ModelProto with output OP deleted

4. CLI Execution

$ sod4onnx \
--input_onnx_file_path movenet_multipose_lightning_192x256_nopost.onnx \
--output_op_names "cast1_output" \
--output_onnx_file_path movenet_multipose_lightning_192x256_nopost_tmp.onnx

5. In-script Execution

from sod4onnx import outputs_delete

onnx_graph = rename(
    input_onnx_file_path="movenet_multipose_lightning_192x256_nopost.onnx",
    output_op_names=["cast1_output"],
    output_onnx_file_path="movenet_multipose_lightning_192x256_nopost_tmp.onnx",
)

6. Sample

$ sod4onnx \
--input_onnx_file_path movenet_multipose_lightning_192x256_nopost.onnx \
--output_op_names "cast1_output" \
--output_onnx_file_path movenet_multipose_lightning_192x256_nopost_tmp.onnx

Before

After

7. Reference

1. https://github.com/onnx/onnx/blob/main/docs/Operators.md
2. https://docs.nvidia.com/deeplearning/tensorrt/onnx-graphsurgeon/docs/index.html
3. https://github.com/NVIDIA/TensorRT/tree/main/tools/onnx-graphsurgeon
4. https://github.com/PINTO0309/simple-onnx-processing-tools
5. https://github.com/PINTO0309/PINTO_model_zoo

8. Issues

https://github.com/PINTO0309/simple-onnx-processing-tools/issues