[Feature]: Add onnx2pplnn tool (open-mmlab#296)

* Add onnx2pplnn tool

* format

* typo

* update opt-shapes and wrapper

* Update opt-shape

* Update comment

* lint

docs/useful_tools.md

@@ -61,6 +61,29 @@ def multiclass_nms(*args, **kwargs):
     """Wrapper function for `_multiclass_nms`."""
 ```
 
+## onnx2pplnn
+
+This tool helps to convert an `ONNX` model to an `PPLNN` model.
+
+### Usage
+
+```bash
+python tools/onnx2pplnn.py \
+    ${ONNX_PATH} \
+    ${OUTPUT_PATH} \
+    --device cuda:0 \
+    --opt-shapes [224,224] \
+    --log-level INFO
+```
+
+### Description of all arguments
+
+- `onnx_path`: The path of the `ONNX` model to convert.
+- `output_path`: The converted `PPLNN` algorithm path in json format.
+- `device`: The device of the model during conversion.
+- `opt-shapes`: Optimal shapes for PPLNN optimization. The shape of each tensor should be wrap with "[]" or "()" and the shapes of tensors should be separated by ",".
+- `--log-level`: To set log level which in `'CRITICAL', 'FATAL', 'ERROR', 'WARN', 'WARNING', 'INFO', 'DEBUG', 'NOTSET'`. If not specified, it will be set to `INFO`.
+
 ## onnx2tensorrt
 
 This tool can be used to convert ONNX to TensorRT engine.

mmdeploy/backend/pplnn/onnx2pplnn.py

@@ -23,8 +23,8 @@ def onnx2pplnn(algo_file: str,
     Args:
         algo_file (str): File path to save PPLNN optimization algorithm.
         onnx_model (str): Input onnx model.
-        device (str): A string specifying cuda device, defaults to 'cuda:0'.
-        input_shapes (Sequence[Sequence[int]] | None): shapes for PPLNN
+        device (str): A string specifying device, defaults to 'cuda:0'.
+        input_shapes (Sequence[Sequence[int]] | None): Shapes for PPLNN
             optimization, default to None.
 
     Examples:

mmdeploy/backend/pplnn/wrapper.py

@@ -189,6 +189,3 @@ def __pplnn_execute(self):
         status = self.runtime.Run()
         assert status == pplcommon.RC_SUCCESS, 'Run() failed: ' + \
             pplcommon.GetRetCodeStr(status)
-        status = self.runtime.Sync()
-        assert status == pplcommon.RC_SUCCESS, 'Sync() failed: ' + \
-            pplcommon.GetRetCodeStr(status)

tools/onnx2pplnn.py

@@ -0,0 +1,68 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import argparse
+import collections
+import logging
+
+from mmdeploy.apis.pplnn import onnx2pplnn
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(description='Convert ONNX to PPLNN.')
+    parser.add_argument('onnx_path', help='ONNX model path')
+    parser.add_argument(
+        'output_path', help='output PPLNN algorithm path in json format')
+    parser.add_argument(
+        '--device',
+        help='`the device of model during conversion',
+        default='cuda:0')
+    parser.add_argument(
+        '--opt-shapes',
+        help='`Optical shapes for PPLNN optimization. The shapes must be able'
+        'to be evaluated by python, e,g., `[1, 3, 224, 224]`',
+        default='[1, 3, 224, 224]')
+    parser.add_argument(
+        '--log-level',
+        help='set log level',
+        default='INFO',
+        choices=list(logging._nameToLevel.keys()))
+    args = parser.parse_args()
+
+    return args
+
+
+def main():
+    args = parse_args()
+    logging.basicConfig(
+        format='%(asctime)s,%(name)s %(levelname)-8s'
+        ' [%(filename)s:%(lineno)d] %(message)s',
+        datefmt='%Y-%m-%d:%H:%M:%S')
+    logger = logging.getLogger()
+    logger.setLevel(args.log_level)
+
+    onnx_path = args.onnx_path
+    output_path = args.output_path
+    device = args.device
+
+    input_shapes = eval(args.opt_shapes)
+    assert isinstance(
+        input_shapes, collections.Sequence), \
+        'The opt-shape must be a sequence.'
+    assert isinstance(input_shapes[0], int) or (isinstance(
+        input_shapes[0], collections.Sequence)), \
+        'The opt-shape must be a sequence of int or a sequence of sequence.'
+    if isinstance(input_shapes[0], int):
+        input_shapes = [input_shapes]
+
+    logging.info(f'onnx2ppl: \n\tonnx_path: {onnx_path} '
+                 f'\n\toutput_path: {output_path}'
+                 f'\n\topt_shapes: {input_shapes}')
+    try:
+        onnx2pplnn(output_path, onnx_path, device, input_shapes)
+        logging.info('onnx2tpplnn success.')
+    except Exception as e:
+        logging.error(e)
+        logging.error('onnx2tpplnn failed.')
+
+
+if __name__ == '__main__':
+    main()