python - Mobilenet from scratch on stanford dogs dataset

Question

Hi I'm trying to teach Mobilenet from scratch on Stanford dogs dataset. My accuracy gets to 0.9 but Val_accuracy to 0.14 what should I do to fix it. I'm not sure if I can use Mobilenet on such dataset

mobilenet(input_shape, n_classes):
def mobilenet_block(x, f, s=1):
    x = DepthwiseConv2D(3, strides=s, padding='same')(x)
    x = BatchNormalization()(x)
    x = ReLU()(x)

    x = Conv2D(f, 1, strides=1, padding='same')(x)
    x = BatchNormalization()(x)
    x = ReLU()(x)
    return x

input = Input(input_shape)

x = Conv2D(32, 3, strides=2, padding='same')(input)
x = BatchNormalization()(x)
x = ReLU()(x)

x = mobilenet_block(x, 64)
x = mobilenet_block(x, 128, 2)
x = mobilenet_block(x, 128)

x = mobilenet_block(x, 256, 2)
x = mobilenet_block(x, 256)

x = mobilenet_block(x, 512, 2)
for _ in range(5):
    x = mobilenet_block(x, 512)

x = mobilenet_block(x, 1024, 2)
x = mobilenet_block(x, 1024)

x = GlobalAvgPool2D()(x)

output = Dense(n_classes, activation='softmax')(x)

model = Model(input, output)
return model

now here is how I'm training the model

enter code here dataset, info = tfds.load(name="stanford_dogs", with_info=True)

get_name = info.features['label'].int2str

IMG_LEN = 224
IMG_SHAPE = (IMG_LEN, IMG_LEN, 3)
N_BREEDS = 120

training_data = dataset['train']
test_data = dataset['test']


model = mobilenet(IMG_SHAPE, 120)

model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
train_batches = prepare(training_data, batch_size=32)
test_batches = prepare(test_data, batch_size=32)

trainedModel = model.fit(train_batches,epochs=30,validation_data=test_batches)
model.save(filename)

predictions = model.predict(prepare_image("images.jpg"))

top_components = tf.reshape(tf.math.top_k(predictions, k=5).indices, shape=[-1])
top_matches = [get_name(i) for i in top_components]

plt.title(top_matches[0])
print(top_matches)

is the way I'm compiling wrong? what am I missing..

Answer 1

等待大神答复