clean code for inference

deploy/models_utils/conversion/onnx2om.sh

@@ -13,6 +13,6 @@ python auto_gear.py --image_path=/xx/lsvt/images \
                     --rec_onnx_path=ch_ppocr_server_v2.0_rec_infer_argmax.onnx \
                     --rec_model_height=32 \
                     --soc_version=Ascend310P \
-                    --output_path=./lsvt_om_v2n &&
+                    --output_path=./lsvt_om_v2 &&
 python auto_select.py --rec_model_path=lsvt_om_v2/crnn
 popd

deploy/mx_infer/args.py

@@ -1,16 +1,20 @@
 import argparse
-import os
 import itertools
+import os
 
-from deploy.mx_infer.utils import log
 from deploy.mx_infer.framework.module_data_type import InferModelComb
 from deploy.mx_infer.processors import SUPPORT_DET_MODEL, SUPPORT_REC_MODEL
+from deploy.mx_infer.utils import log
 
 
 def str2bool(v):
     return v.lower() in ("true", "t", "1")
 
+
 def get_args():
+    """
+    command line parameters for inference
+    """
     parser = argparse.ArgumentParser(description='Arguments for inference.')
     parser.add_argument('--input_images_dir', type=str, required=True,
                         help='Input images dir for inference, can be dir containing multiple images or path of single '
@@ -38,7 +42,7 @@ def get_args():
                         help='Saving dir for visualization of detection results.')
     parser.add_argument('--vis_pipeline_save_dir', type=str, required=False,
                         help='Saving dir for visualization of pipeline inference results.')
-    parser.add_argument('--vis_font_path', type=str, default='', required=False,
+    parser.add_argument('--vis_font_path', type=str, required=False,
                         help='Font file path for recognition model.')
     parser.add_argument('--save_pipeline_crop_res', type=str2bool, default=False, required=False,
                         help='Whether save the images cropped during pipeline.')
@@ -57,13 +61,27 @@ def get_args():
 
 
 def setup_logger(args):
+    """
+    initialize log system
+    """
     log.init_logger(args.show_log, args.save_log_dir)
+    log.save_args(args)
 
 
 def update_task_args(args):
-    det = os.path.exists(args.det_model_path) if isinstance(args.det_model_path, str) else False
-    cls = os.path.exists(args.cls_model_path) if isinstance(args.cls_model_path, str) else False
-    rec = os.path.exists(args.rec_model_path) if isinstance(args.rec_model_path, str) else False
+    """
+    add internal parameters according to different task type
+    """
+    if args.det_model_path and not os.path.exists(args.det_model_path):
+        raise ValueError(f"The det_model_path of '{args.det_model_path}' does not exist.")
+    if args.cls_model_path and not os.path.exists(args.cls_model_path):
+        raise ValueError(f"The cls_model_path of '{args.cls_model_path}' does not exist.")
+    if args.rec_model_path and not os.path.exists(args.rec_model_path):
+        raise ValueError(f"The rec_model_path of '{args.rec_model_path}' does not exist.")
+
+    det = bool(args.det_model_path)
+    cls = bool(args.cls_model_path)
+    rec = bool(args.rec_model_path)
 
     task_map = {
         (True, False, False): InferModelComb.DET,
@@ -74,26 +92,30 @@ def update_task_args(args):
 
     task_order = (det, cls, rec)
     if task_order in task_map:
-        taks_type = task_map[task_order]
-        setattr(args, 'task_type', taks_type)
+        task_type = task_map[task_order]
+        setattr(args, 'task_type', task_type)
+        setattr(args, 'save_vis_det_save_dir', bool(args.vis_det_save_dir))
+        setattr(args, 'save_vis_pipeline_save_dir', bool(args.vis_pipeline_save_dir))
     else:
-        if not (det or cls or rec):
-            raise ValueError(f"det_model_path, cls_model_path, rec_model_path cannot be empty at the same time.")
-        elif det:
-            raise ValueError(f"rec_model_path can't be empty when det_model_path and cls_model_path are not empty.")
-        else:
-            raise ValueError(f"cls_model_path{args.cls_model_path} model does not support inference independently.")
+        unsupported_task_map = {
+            (False, False, False): "empty",
+            (True, True, False): "det+cls",
+            (False, True, False): "cls",
+            (False, True, True): "cls+rec"
+        }
 
-    setattr(args, 'save_vis_det_save_dir', True if args.vis_det_save_dir else False)
-    setattr(args, 'save_vis_pipeline_save_dir', True if args.vis_pipeline_save_dir else False)
+        raise ValueError(
+            f"Only support det, rec, det+rec and det+cls+rec, but got {unsupported_task_map[task_order]}. "
+            f"Please check model_path!")
 
     return args
 
 
 def check_args(args):
-    if not args.input_images_dir or \
-            (not os.path.isfile(args.input_images_dir) and not os.path.isdir(args.input_images_dir)) or \
-            (os.path.isdir(args.input_images_dir) and len(os.listdir(args.input_images_dir)) == 0):
+    """
+    check parameters
+    """
+    if not args.input_images_dir or not os.path.exists(args.input_images_dir):
         raise ValueError(f"input_images_dir must be dir containing multiple images or path of single image.")
 
     if args.det_model_path and not os.path.isfile(args.det_model_path):
@@ -102,8 +124,9 @@ def check_args(args):
     if args.cls_model_path and not os.path.isfile(args.cls_model_path):
         raise ValueError(f"cls_model_path must be a model file path for classification.")
 
-    if args.rec_model_path and (os.path.isdir(args.rec_model_path) and len(os.listdir(args.rec_model_path)) == 0):
-        raise ValueError(f"rec_model_path must a model file or dir containing model file for recognition model.")
+    if args.rec_model_path and (not os.path.exists(args.rec_model_path) or (
+            os.path.isdir(args.rec_model_path) and not os.listdir(args.rec_model_path))):
+        raise ValueError(f"rec_model_path must be a model file or dir containing model file for recognition model.")
 
     if args.rec_model_path and (not args.rec_char_dict_path or not os.path.isfile(args.rec_char_dict_path)):
         raise ValueError(
@@ -127,7 +150,7 @@ def check_args(args):
             f"for single detection task.")
 
     if not args.res_save_dir:
-        raise ValueError(f"res_save_dir can’t be empty.")
+        raise ValueError(f"res_save_dir can't be empty.")
 
     if args.det_algorithm not in SUPPORT_DET_MODEL:
         raise ValueError(f"det_algorithm only support {SUPPORT_DET_MODEL}, but got {args.det_algorithm}.")

deploy/mx_infer/framework/module_base.py

@@ -3,10 +3,9 @@
 from ctypes import c_longdouble
 from multiprocessing import Manager
 
-from .module_data_type import ModuleInitArgs
-
 from deploy.mx_infer.data_type import ProfilingData
 from deploy.mx_infer.utils import log
+from .module_data_type import ModuleInitArgs
 
 
 class ModuleBase(object):
@@ -68,7 +67,7 @@ def process(self, input_data):
         pass
 
     @abstractmethod
-    def init_self_args(self, ):
+    def init_self_args(self):
         self.msg_queue.put(f'{self.__class__.__name__} instance id {self.instance_id} init complete')
         log.info(f'{self.__class__.__name__} instance id {self.instance_id} init complete')
 

deploy/mx_infer/framework/module_manager.py

@@ -1,9 +1,9 @@
 from collections import defaultdict, namedtuple
 from multiprocessing import Queue, Process
 
-from .module_data_type import ModulesInfo, ModuleInitArgs
 from deploy.mx_infer.processors import processor_initiator
 from deploy.mx_infer.utils import log
+from .module_data_type import ModulesInfo, ModuleInitArgs
 
 OutputRegisterInfo = namedtuple('OutputRegisterInfo', ['pipeline_name', 'module_send', 'module_recv'])
 

deploy/mx_infer/pipeline.py

@@ -2,14 +2,19 @@
 import time
 from collections import defaultdict
 from multiprocessing import Process, Queue
+
 import tqdm
+
 from deploy.mx_infer.data_type import StopSign
 from deploy.mx_infer.framework import ModuleDesc, ModuleConnectDesc, ModuleManager, SupportedTaskOrder
 from deploy.mx_infer.processors import MODEL_DICT
 from deploy.mx_infer.utils import log, profiling, safe_div, save_path_init, TASK_QUEUE_SIZE
 
 
 def image_sender(images_path, send_queue, show_progressbar):
+    """
+    send image to input queue for pipeline
+    """
     if os.path.isdir(images_path):
         input_image_list = [os.path.join(images_path, path) for path in os.listdir(images_path)]
         if show_progressbar:
@@ -23,9 +28,12 @@ def image_sender(images_path, send_queue, show_progressbar):
 
 
 def build_pipeline_kernel(args, input_queue):
+    """
+    build and register pipeline
+    """
     task_type = args.task_type
     parallel_num = args.parallel_num
-    module_desc_list = [ModuleDesc('HandoutProcess', 1), ModuleDesc('DecodeProcess', parallel_num), ]
+    module_desc_list = [ModuleDesc('HandoutProcess', 1), ModuleDesc('DecodeProcess', parallel_num)]
 
     module_order = SupportedTaskOrder[task_type]
 
@@ -96,11 +104,15 @@ def build_pipeline(args):
     if args.save_log_dir:
         save_path_init(args.save_log_dir, exist_ok=True)
 
+    if os.path.isdir(args.input_images_dir) and not os.listdir(args.input_images_dir):
+        log.warning(f"The input_images_dir directory '{args.input_images_dir}' is empty, no image to process.")
+        return
+
     task_queue = Queue(TASK_QUEUE_SIZE)
     process = Process(target=build_pipeline_kernel, args=(args, task_queue))
     process.start()
     image_sender(images_path=args.input_images_dir, send_queue=task_queue,
-                 show_progressbar=False if args.show_log else True)
+                 show_progressbar=not args.show_log)
     task_queue.put(StopSign(), block=True)
     process.join()
     process.close()

deploy/mx_infer/processors/__init__.py

@@ -1,8 +1,7 @@
 from sys import modules
 
-from deploy.mx_infer.utils import log
 from deploy.mx_infer.framework import InferModelComb
-
+from deploy.mx_infer.utils import log
 from .classification import CLSPreProcess, CLSInferProcess
 from .common import HandoutProcess, CollectProcess, DecodeProcess
 from .detection import DetPreProcess, DetInferProcess, DetPostProcess, SUPPORT_DET_MODEL

deploy/mx_infer/processors/common/collect_process.py

@@ -9,7 +9,6 @@
 from deploy.mx_infer.data_type import StopData, ProcessData, ProfilingData
 from deploy.mx_infer.framework import ModuleBase, InferModelComb
 from deploy.mx_infer.utils import safe_list_writer, log
-
 from tools.utils.visualize import VisMode, Visualization
 
 _RESULTS_SAVE_FILENAME = {

deploy/mx_infer/processors/detection/det_infer_process.py

@@ -26,7 +26,7 @@ def init_self_args(self):
         if desc != "dynamic_height_width":
             raise ValueError("model input shape must be dynamic image_size with gear.")
 
-        batchsize, channel, hw_list = shape_info
+        _, channel, hw_list = shape_info
         self.gear_list = hw_list
         self.model_channel = channel
         self.max_dot_gear = max([(h, w) for h, w in hw_list], key=lambda x: x[0] * x[1])
@@ -45,7 +45,7 @@ def process(self, input_data):
             self.send_to_next_module(input_data)
             return
         input_array = input_data.input_array
-        n, c, h, w = input_array.shape
+        _, _, h, w = input_array.shape
 
         matched_gear = get_matched_gear_hw((h, w), self.gear_list, self.max_dot_gear)
         input_array = padding_with_np(input_array, matched_gear)

deploy/mx_infer/processors/detection/det_post_process.py

@@ -36,7 +36,7 @@ def get_boxes_from_maps(self, pred: np.ndarray, binary_map: np.ndarray, dest_wid
         outs = cv2.findContours((binary_map * 255).astype(np.uint8), cv2.RETR_LIST,
                                 cv2.CHAIN_APPROX_SIMPLE)
         if len(outs) == 3:
-            img, contours, _ = outs[0], outs[1], outs[2]
+            _, contours, _ = outs[0], outs[1], outs[2]
         elif len(outs) == 2:
             contours, _ = outs[0], outs[1]
 

deploy/mx_infer/processors/recognition/rec_post_process.py

@@ -2,8 +2,8 @@
 
 import numpy as np
 
-from deploy.mx_infer.utils import array_to_texts, file_base_check, log
 from deploy.mx_infer.framework import ModuleBase, InferModelComb
+from deploy.mx_infer.utils import array_to_texts, file_base_check, log
 
 
 class RecPostProcess(ModuleBase):

deploy/mx_infer/utils/cv_utils.py

@@ -93,7 +93,7 @@ def expand(input_images):
     """
     if the type of input_images is numpy array,expand the first axis
     if the type of input_images is list, convert the list to numpy array
-    :param input_images: 
+    :param input_images: input image
     :return: the numpy array of shape (batchsize,channel,height,width)
     """
     if isinstance(input_images, np.ndarray):
@@ -107,9 +107,6 @@ def expand(input_images):
 def unclip(box: np.ndarray, unclip_ratio: float):
     """
     expand the box by unclip ratio
-    :param box: 
-    :param unclip_ratio: 
-    :return: 
     """
     poly = Polygon(box)
     distance = safe_div(poly.area * unclip_ratio, poly.length)
@@ -122,12 +119,6 @@ def unclip(box: np.ndarray, unclip_ratio: float):
 def construct_box(box: np.ndarray, height: int, width: int, dest_height: int, dest_width: int):
     """
     resize the box to the original size.
-    :param box: 
-    :param height: 
-    :param width: 
-    :param dest_height: 
-    :param dest_width: 
-    :return: 
     """
     try:
         box[:, 0] = np.clip(
@@ -175,7 +166,7 @@ def box_score_fast(shrink_map: np.ndarray, input_box: np.ndarray):
     using box mean score as the mean score
     :param shrink_map: the output feature map of DBNet
     :param input_box: the min boxes
-    :return: 
+    :return:
     """
     height, width = shrink_map.shape[:2]
     box = input_box.copy()
@@ -196,7 +187,7 @@ def box_score_slow(shrink_map: np.ndarray, contour: np.ndarray):
     using polyon mean score as the mean score
     :param shrink_map: the output feature map of DBNet
     :param contour: the contours
-    :return: 
+    :return:
     """
     height, width = shrink_map.shape
     contour = contour.copy()
@@ -293,14 +284,14 @@ def get_shape_info(shape: list, gears: list, nchw=True):
     else:
         batchsize, height, width, channel = shape
 
-    if channel != 1 and channel != 3:
+    if channel not in (1, 3):
         raise ValueError("model channel number must be 1 or 3")
 
     # static shape or dynamic shape without gear
     if -1 not in shape:
         return "static_shape", (batchsize, channel, height, width)
 
-    if len(gears) == 0:
+    if not gears:
         return "dynamic_shape", (batchsize, channel, height, width)
 
     # dynamic shape with gear

deploy/mx_infer/utils/logger.py

@@ -177,11 +177,9 @@ def save_args(self, args):
         else:
             logging.error('This api just support argparse or dict, please check your input type.')
             raise Exception('This api just support argparse or dict, please check your input type.')
-        self.debug('Args:')
+        self.info('Args:')
         args_copy = args.copy()
         for key, value in args_copy.items():
-            if isinstance(value, str) and "/" in value:
-                args_copy[key] = ''
             self.info('--> %s: %s', key, self.filter_log_str(args_copy[key]))
         self.info('Finish read param')
 

deploy/mx_infer/utils/safe_utils.py

@@ -126,6 +126,5 @@ def save_path_init(path, exist_ok=False):
     if os.path.exists(path):
         if exist_ok:
             return
-        else:
-            shutil.rmtree(path)
+        shutil.rmtree(path)
     os.makedirs(path, 0o750)

tools/utils/visualize.py

@@ -38,18 +38,14 @@ def vis_bbox(self, image, box_list, color, thickness):
 
     def vis_bbox_text(self, image, box_list, text_list, color, thickness, font_path):
         image_bbox = self.vis_bbox(image, box_list, color, thickness)
-        # image_text = Image.new('RGB', (image_bbox.width, image_bbox.height), (255, 255, 255))
         image_text = image_bbox.copy()
-        # image_text = np.zeros((image_bbox.shape[0], image_bbox.shape[1], 3), np.uint8)
         image_text.fill(255)
         image_text = self.vis_bbox(image_text, box_list, color, thickness)
 
-        # image_bbox = Image.fromarray(cv2.cvtColor(image_bbox, cv2.COLOR_BGR2RGB))
         image_text = Image.fromarray(image_text)
         draw_text = ImageDraw.Draw(image_text)
         font = ImageFont.truetype(font_path, 20, encoding='utf-8')
         for i, text in enumerate(text_list):
-            # draw_text.polygon(box_list[i], fill='blue', outline='blue')
             draw_text.text(box_list[i][0], text, color, font)
         image_concat = np.concatenate([np.array(image_bbox), np.array(image_text)], axis=1)
         return image_concat