• 稍微改了一下，变成我想要的样子
  
  
• 
  
  
  
  
• 
  
  源代码出处  ----> https://github.com/ykwang11/TSN-Tensorflow-unofficial
  
  
• 
  
  
  
  
• 
  
  from __future__ import print_function
  
  
• 
  
  import sys
  
  
• 
  
  import tensorflow as tf
  
  
• 
  
  import numpy as np
  
  
• 
  
  import os
  
  
• 
  
  import time
  
  
• 
  
  import imageio
  
  
• 
  
  import cv2
  
  
• 
  
  
  
  
• 
  
  FLAGS = tf.flags.FLAGS
  
  
• 
  
  tf.flags.DEFINE_boolean('is_training', True, 'Extract the training data or not')
  
  
• 
  
  print('Extracting the training data', FLAGS.is_training)
  
  
• 
  
  # 这些TXT文件可从GitHub获得
  
  
• 
  
  label_set = np.genfromtxt("dataset/ucfTrainTestlist/classInd.txt", dtype ='U')
  
  
• 
  
  training_set = np.genfromtxt("dataset/ucfTrainTestlist/trainlist03.txt", dtype ='U')
  
  
• 
  
  testing_set = np.genfromtxt("dataset/ucfTrainTestlist/testlist03.txt", dtype = 'U')
  
  
• 
  
  
  
  
• 
  
  
  
  
• 
  
  if FLAGS.is_training:
  
  
• 
  
      data_set = training_set[:,0]
  
  
• 
  
  else:
  
  
• 
  
      data_set = testing_set
  
  
• 
  
  print(data_set)
  
  
• 
  
  
  
  
• 
  
  Height = 256
  
  
• 
  
  Width = 340
  
  
• 
  
  
  
  
• 
  
  def make_sure_path_exists(path):
  
  
• 
  
      try:
  
  
• 
  
          os.makedirs(path)
  
  
• 
  
      except OSError:
  
  
• 
  
          if not os.path.isdir(path):
  
  
• 
  
              raise
  
  
• 
  
  
  
  
• 
  
  for data in data_set:
  
  
• 
  
      frame_dir = "F:\Dataset/frame2/" + data.split('.')[0]
  
  
• 
  
      make_sure_path_exists(frame_dir)
  
  
• 
  
  
  
  
• 
  
      file_dir = "F:\Dataset/UCF-101/" + data
  
  
• 
  
      with imageio.get_reader(file_dir,  'ffmpeg') as vid:
  
  
• 
  
          nframes = vid.get_meta_data()['nframes']
  
  
• 
  
          try:
  
  
• 
  
              for i, frame in enumerate(vid):
  
  
• 
  
                  ## note that the order of WH is different between opencv and imagei
  
  
• 
  
                  # o
  
  
• 
  
                  n_frames = i
  
  
• 
  
                  frame = cv2.resize(frame, (Width, Height), interpolation = cv2.INTER_CUBIC)
  
  
• 
  
                  imageio.imwrite(frame_dir+'/frame_%d.jpg' %i, frame)
  
  
• 
  
          except:
  
  
• 
  
              print('runtime error:', data)
  
  
• 
  
              print(n_frames)
  
  
• 
  
          np.save(frame_dir + '/nframes.npy', n_frames)
  
  
  

奈斯，很赞