Please enable JavaScript.

Coggle requires JavaScript to display documents.

InceptionV3 (:question:questions (:five:BN方法如何实现? (:dagger_knife:slim…

- - - - pb
        
        pretrained model
  - - - restore a graph from the file
        
        tf.GraphDef
        
        ParseFromString+read()
        
        import_graph_def
        
        :arrow_right:default graph:question:
        
        :check:
      - eg
        
        with tf.Session() as sess: with tf.gfile.FastGFile(pb_file,'rb') as f: graph_def=tf.GraphDef() graph_def.ParseFromString(f.read()) tf.import_graph_def(graph_def,name='')
        
        :arrow_right:sess.graph
    - - softmax_tensor=sess.graph.get_tensor_by_name('softmax:0')
      - feed
        
        image_data
    - - 2d to 1d
      - sort subscripts based on the value
        
        np.argsort
        
        get final 5
        
        [-5:]
        
        :question:really
        
        :check:
        
        reverse
        
        [::-1]
  - - - python retrain.py \ --bottleneck_dir=bottleneck \ --how_many_training_steps=200 \ --model_dir=inception_model \ --output_graph=retrained_graph.pb \ --output_labels=retrained_labels.txt \ --image_dir=data/flower_photos
      - 我需要准备什么:question:
        
        inception的tgz文件放在model_dir下
        
        图片以文件夹名分类，放在image_dir下
        
        命令行输入retrain命令
    - - 写一个python代码,调用输出的pb和txt文件就行了
        
        $ curl -L https://goo.gl/3lTKZs > label_image.py $ python label_image.py flower_photos/roses/2414954629_3708a1a04d.jpg
- - - - :one:5*5 => 3*3 + 3*3
      - :two:7*7 => 1*7 + 7*1
  - - - :musical_note:跨通道组织信息,提高网络的表达能力,同时可以对输出通道升维及降维,计算量很小却能带来特征变换和非线性化
      - :musical_note:可以在不改变size的情况下修改输出通道数
    - - :explode:卷积层直接和Inception模块组连接.而1个Inception模块组 = Inception Module * n