graph_def.ParseFromString(f.read()) google.protobuf.message.DecodeError：解析消息时出错

如何解决graph_def.ParseFromString(f.read()) google.protobuf.message.DecodeError：解析消息时出错

我试图使用我训练有素的 ssd_resnet50_v1_fpn saved_model.pb 来推断视频，这是由 TensorFlow Object Detection API 2.0 版训练的，我的 TF 版本是 2.4.0。我使用附加的脚本来运行推理，但出现了令人头疼的错误消息：

File "ObjDet_Test_Video.py",line 35,in <module>
    od_graph_def.ParseFromString(serialized_graph)
google.protobuf.message.DecodeError: Error parsing message

下面是我的推理脚本。有大佬知道怎么解决吗？

import numpy as np
import os
import six.moves.urllib as urllib
import sys
import tarfile
import tensorflow as tf
import cv2
import zipfile
from collections import defaultdict
from io import StringIO
from matplotlib import pyplot as plt
from PIL import Image
import numpy as np
import datetime as dt
from objectTracker import * 
from collections import deque
# This is needed since the notebook is stored in the object_detection folder.
sys.path.append("..")
from object_detection.utils import ops as utils_ops

    
from utils import label_map_util
from utils import visualization_utils as vis_util

PATH_TO_CKPT = 'exported-models/my_model/saved_model/saved_model.pb'
PATH_TO_LABELS = os.path.join('annotations','label_map.pbtxt')
NUM_CLASSES = 17

# Load Tensorflow model to memory
detection_graph = tf.Graph()
with detection_graph.as_default():
  od_graph_def = tf.compat.v1.GraphDef() 
  with tf.io.gfile.GFile(PATH_TO_CKPT,'rb') as fid:
    serialized_graph = fid.read()
    od_graph_def.ParseFromString(serialized_graph)
    tf.import_graph_def(od_graph_def,name='')

# Loading label map
label_map = label_map_util.load_labelmap(PATH_TO_LABELS)
categories = label_map_util.convert_label_map_to_categories(label_map,max_num_classes=NUM_CLASSES,use_display_name=True)
category_index = label_map_util.create_category_index(categories)

# Helper code
def load_image_into_numpy_array(image):
  (im_width,im_height) = image.size
  return np.array(image.getdata()).reshape(
      (im_height,im_width,3)).astype(np.uint8)
  
def cvDrawBoundingBox(xLeft,yTop,width,height,bgr,img):
    topLeft = (xLeft,yTop)
    botRight = (xLeft+width,yTop+height)
    #bgr = (0,255)
    thickness = 2
    img = cv2.rectangle(img,topLeft,botRight,thickness)
    return img 

def cvDrawText(txt,x,y,img):    
    botLeft_xy = (x,y)
    font = cv2.FONT_HERSHEY_SIMPLEX
    font_size = 0.66
    color = bgr
    thickness = 1
    lineType = cv2.LINE_AA
    cv2.putText(img,txt,botLeft_xy,font,font_size,color,thickness,lineType)
    return img

def cvDrawTrackBlobs(trackList,img):
    for track in trackList:
        boxColor = (255,255,255)            #white
        if (track.confirmed == True):
            boxColor = (0,255)            #red
        x = track.blob.xLeft
        y = track.blob.yTop
        w = track.blob.width
        h = track.blob.height
        img = cvDrawBoundingBox(x,w,h,boxColor,img)
        blobClass = " "        
        txt = str(track.id)+blobClass
        img = cvDrawText(txt,x-2,y-5,img)
    return img

# Detection

IMAGE_SIZE = (12,8)
with detection_graph.as_default():
  with tf.Session(graph=detection_graph) as sess:
    # Definite input and output Tensors for detection_graph
    image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
    # Each box represents a part of the image where a particular object was detected.
    detection_boxes = detection_graph.get_tensor_by_name('detection_boxes:0')
    # Each score represent how level of confidence for each of the objects.
    # Score is shown on the result image,together with the class label.
    detection_scores = detection_graph.get_tensor_by_name('detection_scores:0')
    detection_classes = detection_graph.get_tensor_by_name('detection_classes:0')
    num_detections = detection_graph.get_tensor_by_name('num_detections:0')    
    cap = cv2.VideoCapture('N_Oakland.avi')
    delay_ms = -1
    #plt.figure(figsize=IMAGE_SIZE)
    #Dictionary for tracks indexed by Id and their state
    trackDict = {id: False for id in range(1,251)}
    trackedObjList = []
    trackTraceQueue = deque([],30)
    frameNum = 0
    while(cap.isOpened()):
          numBlobs = 0          
          blobList = []
          ret,frame = cap.read()
          blobFrame = frame.copy()
          im_height,im_depth = frame.shape
          # Convert open cv color image format (BGR) to RGB as expected by Tensorflow
          image_np = cv2.cvtColor(frame,cv2.COLOR_BGR2RGB)          
          # Expand dimensions since the model expects images to have shape: [1,None,3]
          image_np_expanded = np.expand_dims(image_np,axis=0)
          time_start = dt.datetime.now()
          # Actual detection.
          (boxes,scores,classes,num) = sess.run(
                  [detection_boxes,detection_scores,detection_classes,num_detections],feed_dict={image_tensor: image_np_expanded})
          time_end = dt.datetime.now()
          elapsed_time = time_end - time_start
          # Object Tracking
          numObjs = int(num[0])
          print()
          print("FrameNo:",frameNum)
          print("Objects:",numObjs)
          for idx in range(numObjs):
              if (scores[0][idx] < 0.5):
                  continue
              ymin,xmin,ymax,xmax = boxes[0][idx]
              xmin = int(xmin*im_width)
              xmax = int(xmax*im_width)
              ymin = int(ymin*im_height)
              ymax = int(ymax*im_height)
              objW = xmax-xmin
              objH = ymax-ymin
              #print(idx,ymin,objW,objH,(xmax+xmin)/2,(ymax+ymin)/2,objW*objH)
              #print()
              if (ymax > 0 and objW < im_width/2 and objH < im_height/2):
                  blobList.append(blobAttr(numBlobs,idx,objW*objH))
                  numBlobs += 1
                  #print(idx,objW*objH,scores[0][idx])
              #print(xmin,xmax,ymax)
          """Track moving objects"""
          blobList,trackedObjList,trackDict = trackBlobs(blobList,trackDict)
          blobFrame = cvDrawTrackBlobs(trackedObjList,blobFrame)
          
          # Visualization of the results of a detection.
          vis_util.visualize_boxes_and_labels_on_image_array(
                  image_np,np.squeeze(boxes),np.squeeze(classes).astype(np.int32),np.squeeze(scores),category_index,use_normalized_coordinates=True,line_thickness=2)      
          #plt.imshow(image_np)
          opencv_image=cv2.cvtColor(image_np,cv2.COLOR_RGB2BGR)
          txt = 'Inference Time (secs): '+ str(elapsed_time.total_seconds())
          opencv_image = cvDrawText(txt,10,20,(0,255),opencv_image)
          txt = 'Model: '+ MODEL_NAME[:30]
          opencv_image = cvDrawText(txt,450,opencv_image)
          txt = 'Frame No: '+ str(frameNum)
          blobFrame = cvDrawText(txt,30,0),blobFrame)
          cv2.imshow('Objects',opencv_image)
          cv2.imshow('Tracks',blobFrame)
          frameNum = frameNum + 1
          k = cv2.waitKey(delay_ms) & 0xff          
          if (k == 13): #enter
                if (delay_ms < 0):
                    delay_ms = 30
                else:
                    delay_ms = -1
          if (k == 32): #space
              continue
          elif (k == 27): #Esc 
              break          
    cap.release()
    cv2.destroyAllWindows()