optim some infer yaml, readme and parallel

deploy/py_infer/src/configs/det/ppocr/ch_PP-OCRv2_det_cml.yaml

@@ -1,5 +1,5 @@
 postprocess:
-  name: DistillationDBPostProcess
+  name: DBPostprocess
   binary_thresh: 0.3
   box_thresh: 0.6
   max_candidates: 1000

deploy/py_infer/src/configs/rec/ppocr/arabic_PP-OCRv3_rec.yaml

@@ -1,5 +1,5 @@
 postprocess:
-  name: CTCLabelDecode
+  name: RecCTCLabelDecode
   use_space_char: True
   use_redundant_space_char: True
   blank_at_last: False

deploy/py_infer/src/configs/rec/ppocr/ch_PP-OCRv2_rec_distillation.yaml

@@ -1,5 +1,5 @@
 postprocess:
-  name: DistillationCTCLabelDecode
+  name: RecCTCLabelDecode
   use_space_char: True
   blank_at_last: False
 

deploy/py_infer/src/configs/rec/ppocr/chinese_cht_PP-OCRv3_rec.yaml

@@ -1,5 +1,5 @@
 postprocess:
-  name: CTCLabelDecode
+  name: RecCTCLabelDecode
   use_space_char: True
   use_redundant_space_char: True
   blank_at_last: False

deploy/py_infer/src/configs/rec/ppocr/cyrillic_PP-OCRv3_rec.yaml

@@ -1,5 +1,5 @@
 postprocess:
-  name: CTCLabelDecode
+  name: RecCTCLabelDecode
   use_space_char: True
   use_redundant_space_char: True
   blank_at_last: False

deploy/py_infer/src/configs/rec/ppocr/devanagari_PP-OCRv3_rec.yaml

@@ -1,5 +1,5 @@
 postprocess:
-  name: CTCLabelDecode
+  name: RecCTCLabelDecode
   use_space_char: True
   use_redundant_space_char: True
   blank_at_last: False

deploy/py_infer/src/configs/rec/ppocr/en_PP-OCRv3_rec.yaml

@@ -1,5 +1,5 @@
 postprocess:
-  name: CTCLabelDecode
+  name: RecCTCLabelDecode
   use_space_char: True
   use_redundant_space_char: True
   blank_at_last: False

deploy/py_infer/src/configs/rec/ppocr/japan_PP-OCRv3_rec.yaml

@@ -1,5 +1,5 @@
 postprocess:
-  name: CTCLabelDecode
+  name: RecCTCLabelDecode
   use_space_char: True
   use_redundant_space_char: True
   blank_at_last: False

deploy/py_infer/src/configs/rec/ppocr/ka_PP-OCRv3_rec.yaml

@@ -1,5 +1,5 @@
 postprocess:
-  name: CTCLabelDecode
+  name: RecCTCLabelDecode
   use_space_char: True
   use_redundant_space_char: True
   blank_at_last: False

deploy/py_infer/src/configs/rec/ppocr/korean_PP-OCRv3_rec.yaml

@@ -1,5 +1,5 @@
 postprocess:
-  name: CTCLabelDecode
+  name: RecCTCLabelDecode
   use_space_char: True
   use_redundant_space_char: True
   blank_at_last: False

deploy/py_infer/src/configs/rec/ppocr/latin_PP-OCRv3_rec.yaml

@@ -1,5 +1,5 @@
 postprocess:
-  name: CTCLabelDecode
+  name: RecCTCLabelDecode
   use_space_char: True
   use_redundant_space_char: True
   blank_at_last: False

deploy/py_infer/src/configs/rec/ppocr/rec_chinese_lite_train_v2.0.yaml

@@ -1,5 +1,5 @@
 postprocess:
-  name: CTCLabelDecode
+  name: RecCTCLabelDecode
   use_space_char: True
   blank_at_last: False
 

deploy/py_infer/src/configs/rec/ppocr/rec_en_number_lite_train.yaml

@@ -1,5 +1,5 @@
 postprocess:
-  name: CTCLabelDecode
+  name: RecCTCLabelDecode
   use_space_char: True
   use_redundant_space_char: True
   blank_at_last: False

deploy/py_infer/src/configs/rec/ppocr/ta_PP-OCRv3_rec.yaml

@@ -1,5 +1,5 @@
 postprocess:
-  name: CTCLabelDecode
+  name: RecCTCLabelDecode
   use_space_char: True
   use_redundant_space_char: True
   blank_at_last: False

deploy/py_infer/src/configs/rec/ppocr/te_PP-OCRv3_rec.yaml

@@ -1,5 +1,5 @@
 postprocess:
-  name: CTCLabelDecode
+  name: RecCTCLabelDecode
   use_space_char: True
   use_redundant_space_char: True
   blank_at_last: False

deploy/py_infer/src/data_process/postprocess/postprocess_mapping.py

@@ -4,14 +4,12 @@
     # det
     "DBPostprocess": det_postprocess.DBPostprocess,
     "EASTPostprocess": det_postprocess.EASTPostprocess,
-    "DistillationDBPostProcess": det_postprocess.DBPostprocess,
     "PSEPostprocess": det_postprocess.PSEPostprocess,
     "SASTPostprocess": det_postprocess.SASTPostprocess,
     "FCEPostprocess": det_postprocess.FCEPostprocess,
     # rec
     "RecCTCLabelDecode": rec_postprocess.RecCTCLabelDecode,
-    "CTCLabelDecode": rec_postprocess.RecCTCLabelDecode,
-    "DistillationCTCLabelDecode": rec_postprocess.RecCTCLabelDecode,
+    "RecAttnLabelDecode": rec_postprocess.RecAttnLabelDecode,
     "ViTSTRLabelDecode": rec_postprocess.ViTSTRLabelDecode,
     "AttentionLabelDecode": rec_postprocess.AttentionLabelDecode,
     # cls

deploy/py_infer/src/data_process/postprocess/rec_postprocess.py

@@ -10,6 +10,7 @@
 sys.path.insert(0, mindocr_path)
 
 from mindocr.postprocess import rec_postprocess  # noqa
+from mindocr.postprocess.rec_postprocess import RecAttnLabelDecode  # noqa
 
 
 #  TODO: unify RecCTCLabelDecode with with trained side

deploy/py_infer/src/parallel/datatype/__init__.py

@@ -1,3 +1,3 @@
-from .message_data import ProfilingData, StopSign
+from .message_data import ProfilingData, StartSign, StopSign
 from .module_data import ModuleConnectDesc, ModuleDesc, ModuleInitArgs
 from .process_data import ProcessData, StopData

deploy/py_infer/src/parallel/datatype/message_data.py

@@ -1,6 +1,11 @@
 from dataclasses import dataclass
 
 
+@dataclass
+class StartSign:
+    start: bool = True
+
+
 @dataclass
 class StopSign:
     stop: bool = True

deploy/py_infer/src/parallel/datatype/process_data.py

@@ -18,8 +18,6 @@ class ProcessData:
     image_id: int = ""
     frame: np.ndarray = None
 
-    original_width: int = 0
-    original_height: int = 0
     sub_image_list: list = field(default_factory=lambda: [])
     sub_image_size: int = 0
     data: Union[np.ndarray, List[np.ndarray], Dict] = None
@@ -29,3 +27,4 @@ class ProcessData:
 class StopData:
     skip: bool = True
     image_total: int = 0
+    exception: bool = False

deploy/py_infer/src/parallel/framework/module_base.py

@@ -4,7 +4,7 @@
 from multiprocessing import Manager
 
 from ...utils import log
-from ..datatype import ModuleInitArgs, ProfilingData
+from ..datatype import ModuleInitArgs, ProfilingData, StopData
 
 
 class ModuleBase(object):
@@ -14,7 +14,6 @@ def __init__(self, args, msg_queue):
         self.module_name = ""
         self.without_input_queue = False
         self.instance_id = 0
-        self.device_id = -1
         self.is_stop = False
         self.msg_queue = msg_queue
         self.input_queue = None
@@ -36,9 +35,14 @@ def process_handler(self, stop_manager, input_queue, output_queue):
             log.error(f"{self.__class__.__name__} init failed: {error}")
             raise error
 
-        while not self.msg_queue.full() and stop_manager.full():
+        # waiting for init sign
+        while not self.msg_queue.full():
             continue
-        time.sleep(0.5)
+
+        # waiting for the release of stop sign
+        while stop_manager.full():
+            continue
+
         while True:
             if stop_manager.full():
                 break
@@ -54,7 +58,9 @@ def call_process(self, send_data=None):
             try:
                 self.process(send_data)
             except Exception as error:
+                self.process(StopData(exception=True))
                 log.exception(f"ERROR occurred in {self.module_name} module for {send_data.image_name}: {error}.")
+
             cost_time = time.time() - start_time
             self.process_cost.value += cost_time
 
@@ -85,7 +91,6 @@ def stop(self):
         profiling_data = ProfilingData(
             module_name=self.module_name,
             instance_id=self.instance_id,
-            device_id=self.device_id,
             process_cost_time=self.process_cost.value,
             send_cost_time=self.send_cost.value,
         )

deploy/py_infer/src/parallel/framework/module_manager.py

@@ -12,7 +12,6 @@ class ModuleManager:
     MODULE_QUEUE_MAX_SIZE = 16
 
     def __init__(self, msg_queue: Queue, task_queue: Queue, args):
-        self.device_id = 0
         self.pipeline_map = defaultdict(lambda: defaultdict(ModulesInfo))
         self.msg_queue = msg_queue
         self.stop_manager = Queue(1)
@@ -116,13 +115,6 @@ def run_pipeline(self):
             process.start()
 
     def deinit_pipeline_module(self):
-        # wait for the stop msg
-        while self.stop_manager.empty():
-            continue
-
-        # pop the sign from shared queue
-        self.stop_manager.get()
-
         # the empty() is not reliable, double check the msg queue is empty for receive the profiling data
         while not self.msg_queue.empty():
             self.msg_queue.get()

deploy/py_infer/src/parallel/module/common/collect_node.py

@@ -100,6 +100,8 @@ def process(self, input_data):
             self.result_handle(input_data)
         elif isinstance(input_data, StopData):
             self.stop_handle(input_data)
+            if input_data.exception:
+                self.send_to_next_module("stop")
         else:
             raise ValueError("unknown input data")
 
@@ -111,7 +113,6 @@ def stop(self):
         profiling_data = ProfilingData(
             module_name=self.module_name,
             instance_id=self.instance_id,
-            device_id=self.device_id,
             process_cost_time=self.process_cost.value,
             send_cost_time=self.send_cost.value,
             image_total=self.image_total.value,

deploy/py_infer/src/parallel/module/common/decode_node.py

@@ -7,10 +7,6 @@ def __init__(self, args, msg_queue):
         super().__init__(args, msg_queue)
         self.cost_time = 0
 
-    def decode(self, image_path):
-        image_src = cv_utils.img_read(image_path)
-        return image_src
-
     def init_self_args(self):
         super().init_self_args()
 
@@ -21,11 +17,7 @@ def process(self, input_data):
 
         image_path = input_data.image_path
 
-        image_src = self.decode(image_path)
-        h, w = cv_utils.get_hw_of_img(image_src)
-
+        image_src = cv_utils.img_read(image_path)
         input_data.frame = image_src
-        input_data.original_width = w
-        input_data.original_height = h
 
         self.send_to_next_module(input_data)

deploy/py_infer/src/parallel/parallel_pipeline.py

@@ -8,7 +8,7 @@
 
 from ..infer import SUPPORTED_TASK_BASIC_MODULE
 from ..utils import log, safe_div
-from .datatype import ModuleConnectDesc, ModuleDesc, StopSign
+from .datatype import ModuleConnectDesc, ModuleDesc, StartSign, StopSign
 from .framework import ModuleManager
 from .module import MODEL_DICT
 
@@ -30,6 +30,7 @@ def start_pipeline(self):
         self.input_queue = Queue(self.TASK_QUEUE_SIZE)
         self.process = Process(target=self._build_pipeline_kernel)
         self.process.start()
+        self.input_queue.get(block=True)
 
     def stop_pipeline(self):
         self.input_queue.put(StopSign(), block=True)
@@ -71,20 +72,27 @@ def _build_pipeline_kernel(self):
 
         # waiting for task receive
         while not msg_queue.full():
+            time.sleep(0.1)
             continue
 
-        start_time = time.time()
-        # release all init sign
         for _ in range(module_size):
             msg_queue.get()
 
+        # send start sign for input queue
+        self.input_queue.put(StartSign())
+
         # release the stop sign, infer start
         manager.stop_manager.get(block=False)
 
-        manager.deinit_pipeline_module()
+        start_time = time.time()
+
+        # waiting for inference, and pop the sign from shared queue
+        manager.stop_manager.get(block=True)
 
         cost_time = time.time() - start_time
 
+        manager.deinit_pipeline_module()
+
         # collect the profiling data
         profiling_data = defaultdict(lambda: [0, 0])
         image_total = 0
@@ -95,13 +103,13 @@ def _build_pipeline_kernel(self):
             if msg_info.module_name != -1:
                 image_total = msg_info.image_total
 
-        self.profiling(profiling_data, image_total)
-
-        print(
-            f"Number of images: {image_total}, "
-            f"total cost {cost_time:.2f}s, "
-            f"FPS: {safe_div(image_total, cost_time):.2f}"
-        )
+        if image_total > 0:
+            self.profiling(profiling_data, image_total)
+            print(
+                f"Number of images: {image_total}, "
+                f"total cost {cost_time:.2f}s, "
+                f"FPS: {safe_div(image_total, cost_time):.2f}"
+            )
 
         msg_queue.close()
         msg_queue.join_thread()

docs/cn/inference/models_list_thirdparty.md

@@ -24,7 +24,7 @@ MindOCR可以支持第三方模型（如PaddleOCR、MMOCR等）的推理，本
 | en_pp_det_OCRv3             | DBNet   | MobileNetV3 | PaddleOCR | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/ppocr/ch_PP-OCRv3_det_cml.yaml)                   | [pp-infer](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar)                                                                               | [en_PP-OCRv3_det](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/doc/doc_en/models_list_en.md)               |
 | ml_pp_det_OCRv3             | DBNet   | MobileNetV3 | PaddleOCR | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/ppocr/ch_PP-OCRv3_det_cml.yaml)                   | [pp-infer](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_infer.tar)                                                                | [ml_PP-OCRv3_det](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/doc/doc_en/models_list_en.md)               |
 | en_pp_det_dbnet_resnet50vd  | DBNet   | ResNet50_vd | PaddleOCR | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/ppocr/det_r50_vd_db.yaml)                         | [pp-train](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)                                                                             | [DBNet](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/doc/doc_en/algorithm_det_db_en.md)                    |
-| en_pp_det_psenet_resnet50vd | PSE     | ResNet50_vd | PaddleOCR | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/ppocr/det_r50_vd_pse.yaml)                        | [pp-train](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar)                                                                        | [PSE](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/doc/doc_en/algorithm_overview_en.md)                    |
+| en_pp_det_psenet_resnet50vd | PSE     | ResNet50_vd | PaddleOCR | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/ppocr/det_r50_vd_pse.yaml)                        | [pp-train](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar)                                                                        | [PSE](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/doc/doc_ch/algorithm_det_psenet.md)                     |
 | en_pp_det_east_resnet50vd   | EAST    | ResNet50_vd | PaddleOCR | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/ppocr/det_r50_vd_east.yaml)                       | [pp-train](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)                                                                           | [EAST](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/doc/doc_en/algorithm_det_east_en.md)                   |
 | en_pp_det_sast_resnet50vd   | SAST    | ResNet50_vd | PaddleOCR | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/ppocr/det_r50_vd_sast_icdar15.yaml)               | [pp-train](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar)                                                                   | [SAST](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/doc/doc_en/algorithm_det_sast_en.md)                   |
 | en_mm_det_denetpp_resnet50  | DBNet++ | ResNet50    | MMOCR     | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/mmocr/dbnetpp_resnet50_fpnc_1200e_icdar2015.yaml) | [pth](https://download.openmmlab.com/mmocr/textdet/dbnetpp/dbnetpp_resnet50_fpnc_1200e_icdar2015/dbnetpp_resnet50_fpnc_1200e_icdar2015_20221025_185550-013730aa.pth) | [DBNetpp](https://github.com/open-mmlab/mmocr/blob/main/configs/textdet/dbnetpp/README.md)                               |

docs/en/inference/models_list_thirdparty.md

@@ -27,7 +27,7 @@ The original model files involved are as follows:
 | en_pp_det_OCRv3             | DBNet   | MobileNetV3 | PaddleOCR | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/ppocr/ch_PP-OCRv3_det_cml.yaml)                   | [pp-infer](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar)                                                                               | [en_PP-OCRv3_det](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/doc/doc_en/models_list_en.md)               |
 | ml_pp_det_OCRv3             | DBNet   | MobileNetV3 | PaddleOCR | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/ppocr/ch_PP-OCRv3_det_cml.yaml)                   | [pp-infer](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_infer.tar)                                                                | [ml_PP-OCRv3_det](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/doc/doc_en/models_list_en.md)               |
 | en_pp_det_dbnet_resnet50vd  | DBNet   | ResNet50_vd | PaddleOCR | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/ppocr/det_r50_vd_db.yaml)                         | [pp-train](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)                                                                             | [DBNet](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/doc/doc_en/algorithm_det_db_en.md)                    |
-| en_pp_det_psenet_resnet50vd | PSE     | ResNet50_vd | PaddleOCR | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/ppocr/det_r50_vd_pse.yaml)                        | [pp-train](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar)                                                                        | [PSE](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/doc/doc_en/algorithm_overview_en.md)                    |
+| en_pp_det_psenet_resnet50vd | PSE     | ResNet50_vd | PaddleOCR | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/ppocr/det_r50_vd_pse.yaml)                        | [pp-train](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar)                                                                        | [PSE](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/doc/doc_en/algorithm_det_psenet_en.md)                  |
 | en_pp_det_east_resnet50vd   | EAST    | ResNet50_vd | PaddleOCR | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/ppocr/det_r50_vd_east.yaml)                       | [pp-train](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)                                                                           | [EAST](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/doc/doc_en/algorithm_det_east_en.md)                   |
 | en_pp_det_sast_resnet50vd   | SAST    | ResNet50_vd | PaddleOCR | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/ppocr/det_r50_vd_sast_icdar15.yaml)               | [pp-train](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar)                                                                   | [SAST](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/doc/doc_en/algorithm_det_sast_en.md)                   |
 | en_mm_det_denetpp_resnet50  | DBNet++ | ResNet50    | MMOCR     | [yaml](https://github.com/mindspore-lab/mindocr/tree/main/deploy/py_infer/src/configs/det/mmocr/dbnetpp_resnet50_fpnc_1200e_icdar2015.yaml) | [pth](https://download.openmmlab.com/mmocr/textdet/dbnetpp/dbnetpp_resnet50_fpnc_1200e_icdar2015/dbnetpp_resnet50_fpnc_1200e_icdar2015_20221025_185550-013730aa.pth) | [DBNetpp](https://github.com/open-mmlab/mmocr/blob/main/configs/textdet/dbnetpp/README.md)                               |