神经网络只预测一类

Question

我的模型只从二元类中预测一个类。该模型使用 Keras Video Frame Generator 获取视频输入，并为每个视频获取 350 帧。该模型必须采用 350 帧的输入序列，并使用 BLSTM 输出一个二进制类。输入形状为 (350, 112, 75, 3)。由于 OOM 错误，批大小为 2。我不知道这可能是问题还是代码中有问题，但模型似乎什么也没学到。这是代码：

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Conv2D, BatchNormalization, MaxPool2D, GlobalMaxPool2D
def build_convnet(shape=(112, 75, 2)):
    momentum = .9
    model = Sequential()
    model.add(Conv2D(64, (3,3), input_shape=shape, padding='same', activation='relu'))
    model.add(Conv2D(64, (3,3), padding='same', activation='relu'))
    model.add(BatchNormalization(momentum=momentum))
    
    model.add(MaxPool2D())
    
    model.add(Conv2D(128, (3,3), padding='same', activation='relu'))
    model.add(Conv2D(128, (3,3), padding='same', activation='relu'))
    model.add(BatchNormalization(momentum=momentum))
    
    model.add(MaxPool2D())
    
    model.add(Conv2D(256, (3,3), padding='same', activation='relu'))
    model.add(Conv2D(256, (3,3), padding='same', activation='relu'))
    model.add(BatchNormalization(momentum=momentum))
    
    model.add(MaxPool2D())
    
    model.add(Conv2D(512, (3,3), padding='same', activation='relu'))
    model.add(Conv2D(512, (3,3), padding='same', activation='relu'))
    model.add(BatchNormalization(momentum=momentum))
    
    # flatten...
    model.add(GlobalMaxPool2D())
    
    return model

from tensorflow.keras.layers import LSTM, Bidirectional
from tensorflow.keras.layers import TimeDistributed, Dense, Dropout
def action_model(shape=(350, 112, 75, 3), nbout=2):
    # Create our convnet with (112, 75, 3) input shape
    convnet = build_convnet(shape[1:])
    
    # then create our final model
    model = Sequential()
    
    # add the convnet with (350, 112, 75, 3) shape
    model.add(TimeDistributed(convnet, input_shape=shape))
    model.add(Bidirectional(LSTM(units = 512, return_sequences = True, input_shape = (NBFRAME, 112*75*3))))
    model.add(Dropout(0.5))

    # Adding a second LSTM layer and Dropout layer
    model.add(Bidirectional(LSTM(units = 512, return_sequences = True)))
    model.add(Dropout(0.5))

    # Adding a third LSTM layer and Dropout layer
    model.add(Bidirectional(LSTM(units = 512)))
    model.add(Dropout(0.5))
    model.add(Dense(nbout, activation='softmax'))
    model.summary()
    return model

这是模型摘要。

Answer 1

我认为您在内存中放入了大量数据，您是否尝试过减少视频的帧数？这有帮助吗？

Answer 2

你的模型太大了，我把这块去掉：

emodel.add(Conv2D(128, (3,3), padding='same', activation='relu'))
model.add(Conv2D(128, (3,3), padding='same', activation='relu')) 
model.add(BatchNormalization(momentum=momentum)) 
model.add(MaxPool2D())         
model.add(Conv2D(256, (3,3), padding='same', activation='relu')) 
model.add(Conv2D(256, (3,3), padding='same', activation='relu'))   
model.add(BatchNormalization(momentum=momentum))          
model.add(MaxPool2D())       
model.add(Conv2D(512, (3,3), padding='same', activation='relu'))  
model.add(Conv2D(512, (3,3), padding='same', activation='relu')) 
model.add(BatchNormalization(momentum=momentum))

也许它会导致 oom，也许你不需要它。