在 Tensorflow Keras 中运行 model.fit 时出现 NotImplementedError

Question

我目前正在研究 Aeriel 视频上的人类行为识别。我正在使用这个dataset。您可以看到videos 和labels 文件。我正在构建一个 SSD 模型来训练数据..

我在使用 model.fit 时遇到错误。

我认为主要问题出在 DataGenerator 类中，尽管我无法解决该错误。代码如下

import numpy as np
import cv2
from tensorflow.keras.utils import Sequence
import tensorflow as tf
import os
import json


class DataGenerator(tf.keras.utils.Sequence):
   def __init__(self, img_dir, ann_dir, 
                batch_size=32, dim=(300,300),
                shuffle=True):
       self.idx_to_name = ["None", '"Handshaking"\n', '"Hugging"\n', '"Reading"\n', '"Drinking"\n', '"Pushing/Pulling"\n', '"Carrying"\n', '"Calling"\n', '"Running"\n', '"Walking"\n', '"Lying"\n', '"Sitting"\n', '"Standing"\n']
       self.name_to_idx = dict([(v, k) for k, v in enumerate(self.idx_to_name)])
       self.img_dir = img_dir
       self.ann_dir = ann_dir
       # self.frame = frame
       self.batch_size = batch_size
       self.dim = dim



   def _get_annotation(self, file, j):
       
       frame_map = dict()
       with open(file, 'r') as fp:
           line = fp.readline()

       while line:
           line_split = line.split(' ')
           frame_id = int(line_split[5])
           if line_split[10] is not None:
               label = line_split[10]
           else:
               label = "None"
           val = (int(line_split[0]), list(map(int, line_split[1:5])), list(map(int, line_split[6:8])), line_split[10])
           if frame_id not in frame_map:
               frame_map[frame_id] = [val]
           else:
               frame_map[frame_id].append(val)

           line = fp.readline()
   
       for obj in frame_map[int(j)]:
           xmin = float(obj[1][0]) / 3840.0
           ymin = float(obj[1][1]) / 2160.0
           xmax = float(obj[1][2]) / 3840.0
           ymax = float(obj[1][3]) / 2160.0
           name = obj[3]
           boxes.append([xmin, ymin, xmax, ymax])

           labels.append(self.name_to_idx[name] + 1)

       return np.array(boxes, dtype=np.float32), np.array(labels, dtype=np.int64)

   def __getitem__(self, index):
       start_index = index * self.batch_size
       x_train = []
       y_train = []
       i = start_index - 1
       while len(x_train) < self.batch_size:
           try:
               for i in os.listdir(self.img_dir):
                 for j in os.listdir(self.img_dir + '/' + i):
                   img = cv2.imread(self.img_dir + '/' +i + '/' + j)
                   img = cv2.resize(img,(320,240))
                   img = np.array(img, dtype = np.float32)
                   img = img / 255.0
                   boxes, labels = self._get_annotation(self.ann_dir + '/' + i + '.txt', int(j[:-4]))
                   x_train.append(img)
                   y_train.append(boxes)
                   i += 1

           except Exception as err:
               print(err)
               continue
           
       x_train = np.array(x_train)
       y_train = np.array(y_train)
 
       return x_train, y_train

train_data = DataGenerator("/content/okutama_imgs", "/content/okutama_labels", batch_size=4)

model.compile(optimizer=optimizer, loss= SSD_loss, metrics=['accuracy'])
model.fit(train_data, epochs = 50, verbose = 1, callbacks = callbacks)

运行 model.fit 时出现此错误。我不明白为什么会这样，如果您需要更多信息，我很乐意提供给您。这是错误

NotImplementedError                       Traceback (most recent call last)
<ipython-input-64-7c278c6b3232> in <module>()
----> 1 model.fit(train_data, epochs = 50, verbose = 1, callbacks = callbacks)

4 frames
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/utils/data_utils.py in __len__(self)
    456         The number of batches in the Sequence.
    457     """
--> 458     raise NotImplementedError
    459 
    460   def on_epoch_end(self):

NotImplementedError: 

Answer 1

这个错误表明你没有在继承的类中实现一些需要的函数，在这种情况下是__len__函数。

像这样将__len__ 函数添加到您的DataGenerator 类中：

def __len__(self):
    #len(self.x) is the length of your input features
    return math.ceil(len(self.x) / self.batch_size)