Support mmocr:dev-1.x (open-mmlab#904)

* init

* update UT

* fix UT except SAR

* update to latest 2.0

* fix ncnn UT

* export info

.gitignore

@@ -153,6 +153,8 @@ mmdeploy/backend/ncnn/onnx2ncnn
 
 /mmdeploy-*
 
+# OCR dicts
+dicts
 # snpe
 grpc-cpp-plugin
 service/snpe/grpc_cpp_plugin

configs/mmocr/text-detection/text-detection_sdk_dynamic.py

@@ -4,5 +4,5 @@
 
 backend_config = dict(pipeline=[
  dict(type='LoadImageFromFile'),
- dict(type='Collect', keys=['img'], meta_keys=['filename', 'ori_shape'])
+ dict(type='PackTextDetInputs', meta_keys=['img_path', 'ori_shape'])
 ])

mmdeploy/apis/__init__.py

@@ -1,6 +1,6 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 
-# mmcv dependency
+# mmcv & mmengine dependency
 try:
  from .calibration import create_calib_input_data
  from .extract_model import extract_model

mmdeploy/apis/inference.py

@@ -1,15 +1,15 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 from typing import Any, Sequence, Union
 
-import mmcv
+import mmengine
 import numpy as np
 import torch
 
 from mmdeploy.utils import get_input_shape, load_config
 
 
-def inference_model(model_cfg: Union[str, mmcv.Config],
- deploy_cfg: Union[str, mmcv.Config],
+def inference_model(model_cfg: Union[str, mmengine.Config],
+ deploy_cfg: Union[str, mmengine.Config],
  backend_files: Sequence[str], img: Union[str, np.ndarray],
  device: str) -> Any:
  """Run inference with PyTorch or backend model and show results.
@@ -27,8 +27,8 @@ def inference_model(model_cfg: Union[str, mmcv.Config],
  backend_files, img, device)
 
  Args:
- model_cfg (str | mmcv.Config): Model config file or Config object.
- deploy_cfg (str | mmcv.Config): Deployment config file or Config
+ model_cfg (str | mmengine.Config): Model config file or Config object.
+ deploy_cfg (str | mmengine.Config): Deployment config file or Config
  object.
  backend_files (Sequence[str]): Input backend model file(s).
  img (str | np.ndarray): Input image file or numpy array for inference.

mmdeploy/apis/openvino/utils.py

@@ -1,7 +1,7 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 from typing import Dict, List
 
-import mmcv
+import mmengine
 
 from mmdeploy.backend.openvino import ModelOptimizerOptions
 from mmdeploy.utils import get_model_inputs
@@ -30,12 +30,12 @@ def update_input_names(input_info: Dict[str, List],
  return input_info
 
 
-def get_input_info_from_cfg(deploy_cfg: mmcv.Config) -> Dict[str, List]:
+def get_input_info_from_cfg(deploy_cfg: mmengine.Config) -> Dict[str, List]:
  """Get the input names and shapes from the configs for OpenVINO Model
  Optimizer.
 
  Args:
- deploy_cfg (mmcv.Config): Deployment config.
+ deploy_cfg (mmengine.Config): Deployment config.
 
  Returns:
  Dict[str, List]: A dict that stores the names and shapes of input.
@@ -53,12 +53,13 @@ def get_input_info_from_cfg(deploy_cfg: mmcv.Config) -> Dict[str, List]:
  return input_info
 
 
-def get_mo_options_from_cfg(deploy_cfg: mmcv.Config) -> ModelOptimizerOptions:
+def get_mo_options_from_cfg(
+ deploy_cfg: mmengine.Config) -> ModelOptimizerOptions:
  """Get additional parameters for the Model Optimizer from the deploy
  config.
 
  Args:
- deploy_cfg (mmcv.Config): Deployment config.
+ deploy_cfg (mmengine.Config): Deployment config.
 
  Returns:
  ModelOptimizerOptions: A class that will contain additional arguments.

mmdeploy/apis/pytorch2onnx.py

@@ -2,7 +2,7 @@
 import os.path as osp
 from typing import Any, Optional, Union
 
-import mmcv
+import mmengine
 import torch
 
 from mmdeploy.apis.core.pipeline_manager import no_mp
@@ -16,8 +16,8 @@
 def torch2onnx(img: Any,
  work_dir: str,
  save_file: str,
- deploy_cfg: Union[str, mmcv.Config],
- model_cfg: Union[str, mmcv.Config],
+ deploy_cfg: Union[str, mmengine.Config],
+ model_cfg: Union[str, mmengine.Config],
  model_checkpoint: Optional[str] = None,
  device: str = 'cuda:0'):
  """Convert PyTorch model to ONNX model.
@@ -42,16 +42,16 @@ def torch2onnx(img: Any,
  converting model.
  work_dir (str): A working directory to save files.
  save_file (str): Filename to save onnx model.
- deploy_cfg (str | mmcv.Config): Deployment config file or
+ deploy_cfg (str | mmengine.Config): Deployment config file or
  Config object.
- model_cfg (str | mmcv.Config): Model config file or Config object.
+ model_cfg (str | mmengine.Config): Model config file or Config object.
  model_checkpoint (str): A checkpoint path of PyTorch model,
  defaults to `None`.
  device (str): A string specifying device type, defaults to 'cuda:0'.
  """
  # load deploy_cfg if necessary
  deploy_cfg, model_cfg = load_config(deploy_cfg, model_cfg)
- mmcv.mkdir_or_exist(osp.abspath(work_dir))
+ mmengine.mkdir_or_exist(osp.abspath(work_dir))
 
  input_shape = get_input_shape(deploy_cfg)
 
@@ -66,7 +66,7 @@ def torch2onnx(img: Any,
  data_preprocessor=getattr(torch_model, 'data_preprocessor', None))
  if not isinstance(model_inputs, torch.Tensor) and len(model_inputs) == 1:
  model_inputs = model_inputs[0]
- data_samples = data[1]
+ data_samples = data['data_samples']
  patch_metas = {'data_samples': data_samples}
  input_metas = {'data_samples': data_samples, 'mode': 'predict'}
 

mmdeploy/apis/pytorch2torchscript.py

@@ -2,7 +2,7 @@
 import os.path as osp
 from typing import Any, Optional, Union
 
-import mmcv
+import mmengine
 import torch
 
 from mmdeploy.apis.core.pipeline_manager import PIPELINE_MANAGER, no_mp
@@ -14,8 +14,8 @@
 def torch2torchscript(img: Any,
  work_dir: str,
  save_file: str,
- deploy_cfg: Union[str, mmcv.Config],
- model_cfg: Union[str, mmcv.Config],
+ deploy_cfg: Union[str, mmengine.Config],
+ model_cfg: Union[str, mmengine.Config],
  model_checkpoint: Optional[str] = None,
  device: str = 'cuda:0'):
  """Convert PyTorch model to torchscript model.
@@ -25,16 +25,16 @@ def torch2torchscript(img: Any,
  converting model.
  work_dir (str): A working directory to save files.
  save_file (str): Filename to save torchscript model.
- deploy_cfg (str | mmcv.Config): Deployment config file or
+ deploy_cfg (str | mmengine.Config): Deployment config file or
  Config object.
- model_cfg (str | mmcv.Config): Model config file or Config object.
+ model_cfg (str | mmengine.Config): Model config file or Config object.
  model_checkpoint (str): A checkpoint path of PyTorch model,
  defaults to `None`.
  device (str): A string specifying device type, defaults to 'cuda:0'.
  """
  # load deploy_cfg if necessary
  deploy_cfg, model_cfg = load_config(deploy_cfg, model_cfg)
- mmcv.mkdir_or_exist(osp.abspath(work_dir))
+ mmengine.mkdir_or_exist(osp.abspath(work_dir))
 
  input_shape = get_input_shape(deploy_cfg)
 

mmdeploy/apis/utils/calibration.py

@@ -38,10 +38,10 @@ def create_calib_input_data(calib_file: str,
 
  Args:
  calib_file (str): Input calibration file.
- deploy_cfg (str | mmcv.Config): Deployment config.
- model_cfg (str | mmcv.Config): The model config.
+ deploy_cfg (str | mmengine.Config): Deployment config.
+ model_cfg (str | mmengine.Config): The model config.
  model_checkpoint (str): PyTorch model checkpoint, defaults to `None`.
- dataset_cfg (str | mmcv.Config): Dataset config, defaults to `None`
+ dataset_cfg (str | mmengine.Config): Dataset config, defaults to `None`
  dataset_type (str): A string specifying dataset type, e.g.: 'test',
  'val', defaults to 'val'.
  device (str): Specifying the device to run on, defaults to 'cpu'.

mmdeploy/apis/utils/utils.py

@@ -1,17 +1,17 @@
 # Copyright (c) OpenMMLab. All rights reserved.
-import mmcv
+import mmengine
 
 from mmdeploy.codebase import BaseTask, get_codebase_class, import_codebase
 from mmdeploy.utils import get_codebase, get_task_type
 
 
-def build_task_processor(model_cfg: mmcv.Config, deploy_cfg: mmcv.Config,
- device: str) -> BaseTask:
+def build_task_processor(model_cfg: mmengine.Config,
+ deploy_cfg: mmengine.Config, device: str) -> BaseTask:
  """Build a task processor to manage the deployment pipeline.
 
  Args:
- model_cfg (str | mmcv.Config): Model config file.
- deploy_cfg (str | mmcv.Config): Deployment config file.
+ model_cfg (str | mmengine.Config): Model config file.
+ deploy_cfg (str | mmengine.Config): Deployment config file.
  device (str): A string specifying device type.
 
  Returns:
@@ -23,14 +23,15 @@ def build_task_processor(model_cfg: mmcv.Config, deploy_cfg: mmcv.Config,
  return codebase.build_task_processor(model_cfg, deploy_cfg, device)
 
 
-def get_predefined_partition_cfg(deploy_cfg: mmcv.Config, partition_type: str):
+def get_predefined_partition_cfg(deploy_cfg: mmengine.Config,
+ partition_type: str):
  """Get the predefined partition config.
 
  Notes:
  Currently only support mmdet codebase.
 
  Args:
- deploy_cfg (mmcv.Config): use deploy config to get the codebase and
+ deploy_cfg (mmengine.Config): use deploy config to get the codebase and
  task type.
  partition_type (str): A string specifying partition type.
 

mmdeploy/apis/visualize.py

@@ -1,16 +1,16 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 from typing import Optional, Sequence, Union
 
-import mmcv
+import mmengine
 import numpy as np
 import torch
 
 from mmdeploy.codebase import BaseTask
 from mmdeploy.utils import Backend, get_backend, get_input_shape, load_config
 
 
-def visualize_model(model_cfg: Union[str, mmcv.Config],
- deploy_cfg: Union[str, mmcv.Config],
+def visualize_model(model_cfg: Union[str, mmengine.Config],
+ deploy_cfg: Union[str, mmengine.Config],
  model: Union[str, Sequence[str], BaseTask],
  img: Union[str, np.ndarray],
  device: str,
@@ -33,8 +33,8 @@ def visualize_model(model_cfg: Union[str, mmcv.Config],
  img, device, show_result=True)
 
  Args:
- model_cfg (str | mmcv.Config): Model config file or Config object.
- deploy_cfg (str | mmcv.Config): Deployment config file or Config
+ model_cfg (str | mmengine.Config): Model config file or Config object.
+ deploy_cfg (str | mmengine.Config): Deployment config file or Config
  object.
  model (str | list[str], BaseSubtask): Input model or file(s).
  img (str | np.ndarray): Input image file or numpy array for inference.
@@ -69,7 +69,7 @@ def visualize_model(model_cfg: Union[str, mmcv.Config],
 
  model_inputs, _ = task_processor.create_input(img, input_shape)
  with torch.no_grad():
- result = model.test_step([model_inputs])[0]
+ result = model.test_step(model_inputs)[0]
 
  task_processor.visualize(
  image=img,

mmdeploy/backend/base/backend_wrapper_registry.py

@@ -1,5 +1,5 @@
 # Copyright (c) OpenMMLab. All rights reserved.
-from mmcv.utils import Registry
+from mmengine import Registry
 
 from mmdeploy.utils.config_utils import Backend
 

mmdeploy/backend/ncnn/quant.py

@@ -3,7 +3,7 @@
 from subprocess import call
 from typing import List
 
-import mmcv
+import mmengine
 
 from .init_plugins import get_ncnn2int8_path
 
@@ -19,7 +19,7 @@ def get_quant_model_file(onnx_path: str, work_dir: str) -> List[str]:
  List[str]: The path to the files where the export result will be
  located.
  """
- mmcv.mkdir_or_exist(osp.abspath(work_dir))
+ mmengine.mkdir_or_exist(osp.abspath(work_dir))
  base_name = osp.splitext(osp.split(onnx_path)[1])[0]
  quant_onnx = osp.join(work_dir, base_name + '_quant.onnx')
  quant_table = osp.join(work_dir, base_name + '.table')

mmdeploy/backend/openvino/onnx2openvino.py

@@ -5,7 +5,7 @@
 from subprocess import PIPE, CalledProcessError, run
 from typing import Dict, Optional, Sequence, Union
 
-import mmcv
+import mmengine
 import onnx
 
 from mmdeploy.utils import get_root_logger
@@ -50,7 +50,7 @@ def get_output_model_file(onnx_path: str, work_dir: str) -> str:
  Returns:
  str: The path to the file where the export result will be located.
  """
- mmcv.mkdir_or_exist(osp.abspath(work_dir))
+ mmengine.mkdir_or_exist(osp.abspath(work_dir))
  file_name = osp.splitext(osp.split(onnx_path)[1])[0]
  model_xml = osp.join(work_dir, file_name + '.xml')
  return model_xml

mmdeploy/backend/pplnn/utils.py

@@ -50,6 +50,7 @@ def register_engines(device_id: int,
  export_algo_file: str = None,
  import_algo_file: str = None) -> List[pplnn.Engine]:
  """Register engines for pplnn runtime.
+
  Args:
  device_id (int): Specifying device index. `-1` for cpu.
  disable_avx512 (bool): Whether to disable avx512 for x86.

mmdeploy/backend/pplnn/wrapper.py

@@ -68,6 +68,7 @@ def __init__(self,
  def forward(self, inputs: Dict[str,
  torch.Tensor]) -> Dict[str, torch.Tensor]:
  """Run forward inference.
+
  Args:
  inputs (Dict[str, torch.Tensor]): Input name and tensor pairs.
  Return: