Keras: CNN classifier multiclass

Question

Après avoir commencé avec l'exemple de classification binaire officiel de Keras (voir here), j'implémente un classificateur multiclasses avec Tensorflow comme backend. Dans cet exemple, il y a deux classes (chien/chat), j'ai maintenant 50 classes, et les données sont stockées de la même manière dans les dossiers.

Lors de l'entraînement, la perte ne diminuera pas et la précision ne montera pas. J'ai changé la dernière couche qui a utilisé une fonction sigmoid pour utiliser le softmax, a changé binary_crossentropy en categorical_crossentropy, et a changé le class_mode en categorical.

Voici mon code:

from keras.preprocessing.image import ImageDataGenerator 
from keras.models import Sequential 
from keras.layers import Conv2D, MaxPooling2D 
from keras.layers import Activation, Dropout, Flatten, Dense 
from keras import backend as K 
import keras.optimizers 



optimizer = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True) 

# dimensions of our images. 
img_width, img_height = 224, 224 

train_data_dir = 'images/train' 
validation_data_dir = 'images/val' 
nb_train_samples = 209222 
nb_validation_samples = 40000 
epochs = 50 
batch_size = 16 

if K.image_data_format() == 'channels_first': 
    input_shape = (3, img_width, img_height) 
else: 
    input_shape = (img_width, img_height, 3) 

model = Sequential() 
model.add(Conv2D(32, (3, 3), input_shape=input_shape)) 
model.add(Activation('relu')) 
model.add(MaxPooling2D(pool_size=(2, 2))) 

model.add(Conv2D(32, (3, 3))) 
model.add(Activation('relu')) 
model.add(MaxPooling2D(pool_size=(2, 2))) 

model.add(Conv2D(64, (3, 3))) 
model.add(Activation('relu')) 
model.add(MaxPooling2D(pool_size=(2, 2))) 

model.add(Flatten()) 
model.add(Dense(64)) 
model.add(Activation('relu')) 
model.add(Dropout(0.5)) 
model.add(Dense(50)) 
model.add(Activation('softmax')) 



model.compile(loss='categorical_crossentropy', optimizer=optimizer, metrics=['accuracy']) 


train_datagen = ImageDataGenerator() 

train_generator = train_datagen.flow_from_directory(
    directory=train_data_dir, 
    target_size=(img_width, img_height), 
    batch_size=batch_size, 
    class_mode='categorical') 

validation_generator = train_datagen.flow_from_directory(
    directory=validation_data_dir, 
    target_size=(img_width, img_height), 
    batch_size=batch_size, 
    class_mode='categorical') 

model.fit_generator(
    train_generator, 
    steps_per_epoch=nb_train_samples // batch_size, 
    epochs=epochs, 
    validation_data=validation_generator, 
    validation_steps=nb_validation_samples // batch_size) 

model.save_weights('weights.h5') 

Toute idée sur l'endroit où je peux me tromper? Toute contribution sera grandement appréciée!

EDIT: comme demandé par @RobertValencia, voici le début des derniers journaux de formation:

Using TensorFlow backend. 
I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcublas.so.7.5 locally 
I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcudnn.so.5 locally 
I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcufft.so.7.5 locally 
I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcuda.so.1 locally 
I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcurand.so.7.5 locally 
Found 3517 images belonging to 50 classes. 
<keras.preprocessing.image.DirectoryIterator object at 0x7fd1d4515c10> 
Found 2451 images belonging to 50 classes. 
Epoch 1/50 
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE3 instructions, but these are available on your machine and could speed up CPU computations. 
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations. 
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations. 
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations. 
I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:910] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero 
I tensorflow/core/common_runtime/gpu/gpu_device.cc:885] Found device 0 with properties: 
name: GRID K520 
major: 3 minor: 0 memoryClockRate (GHz) 0.797 
pciBusID 0000:00:03.0 
Total memory: 3.94GiB 
Free memory: 3.91GiB 
I tensorflow/core/common_runtime/gpu/gpu_device.cc:906] DMA: 0 
I tensorflow/core/common_runtime/gpu/gpu_device.cc:916] 0: Y 
I tensorflow/core/common_runtime/gpu/gpu_device.cc:975] Creating TensorFlow device (/gpu:0) -> (device: 0, name: GRID K520, pci bus id: 0000:00:03.0) 
8098/13076 [=================>............] - ETA: 564s - loss: 15.6869 - categorical_accuracy: 0.0267 

Answer 1

Compte tenu du nombre de classes que vous avez besoin de faire la différence, ce qui augmente peut-être la complexité du modèle, ainsi en utilisant un optimiseur différent, pourrait donner de meilleurs résultats. Essayez d'utiliser ce modèle, qui est en partie basé sur la VGG-16 l'architecture CNN, mais pas aussi complexe:

model = Sequential() 
model.add(Convolution2D(32, 3, 3, activation='relu')) 
model.add(Convolution2D(32, 3, 3, activation='relu')) 
model.add(MaxPooling2D((2,2), strides=(2,2))) 

model.add(Convolution2D(64, 3, 3, activation='relu')) 
model.add(Convolution2D(64, 3, 3, activation='relu')) 
model.add(MaxPooling2D((2,2), strides=(2,2))) 

model.add(Convolution2D(128, 3, 3, activation='relu')) 
model.add(Convolution2D(128, 3, 3, activation='relu')) 
model.add(MaxPooling2D((2,2), strides=(2,2))) 

model.add(Convolution2D(256, 3, 3, activation='relu')) 
model.add(Convolution2D(256, 3, 3, activation='relu')) 
model.add(MaxPooling2D((2,2), strides=(2,2))) 

model.add(Convolution2D(512, 3, 3, activation='relu')) 
model.add(Convolution2D(512, 3, 3, activation='relu')) 
model.add(MaxPooling2D((2,2), strides=(2,2))) 

model.add(Flatten()) 
model.add(Dense(1024, activation='relu')) 
model.add(Dense(1024, activation='relu')) 
model.add(Dense(50, activation='softmax')) 

optimizer = Nadam(lr=0.002, 
        beta_1=0.9, 
        beta_2=0.999, 
        epsilon=1e-08, 
        schedule_decay=0.004) 

model.compile(loss='categorical_crossentropy', 
       optimizer=optimizer, 
       metrics=['categorical_accuracy']) 

Si vous obtenez de meilleurs résultats, je suggère à la recherche dans le modèle VGG-16:

1. https://github.com/fchollet/keras/blob/master/keras/applications/vgg16.py ou
2. https://gist.github.com/baraldilorenzo/07d7802847aaad0a35d3 (y compris zéro rembourrage et couches d'abandon)