CODEWITHSUNDEEP
pythonmachine-learningtensorflowkerascaptcha
cnian
cnian

Reputation: 249

Low accuracy on captcha recognition

I am generating at most 4 digits captchas with the following method:

def genData(n=30000, max_digs=4, width=150):
    capgen = ImageCaptcha()
    data = []
    target = []
    for i in range(n):
        x = np.random.randint(0, 10 ** max_digs)
        img = misc.imread(capgen.generate(str(x)))
        img = np.mean(img, axis=2)[:, :width]
        data.append(img.flatten())
        target.append(x)
    return np.array(data), np.array(target)

Then I am processing data like following

train_data, train_target = genData()
test_data, test_target = genData(1000)

train_data = train_data.reshape(train_data.shape[0], 1, 150, 60)
test_data = test_data.reshape(test_data.shape[0], 1, 150, 60)
train_data = train_data.astype('float32')
test_data = test_data.astype('float32')
train_data /= 255
test_data /= 255

My model structure is as follows:

def get_model():
    # create model
    model = Sequential()
    model.add(Conv2D(30, (5, 5), input_shape=(1, 150, 60), activation='relu'))
    model.add(MaxPooling2D(pool_size=(2, 2)))
    model.add(Conv2D(15, (3, 3), activation='relu'))
    model.add(MaxPooling2D(pool_size=(2, 2)))
    model.add(Dropout(0.2))
    model.add(Flatten())
    model.add(Dense(128, activation='relu'))
    model.add(Dense(50, activation='relu'))
    model.add(Dense(10 ** 4, activation='softmax'))
    # Compile model
    model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
    return model

Then I am training the model 

model = get_model()
# Fit the model
model.fit(train_data, train_target, validation_data=(test_data, test_target), epochs=10, batch_size=200)
# Final evaluation of the model
scores = model.evaluate(test_data, test_target, verbose=0)
print("Large CNN Error: %.2f%%" % (100 - scores[1] * 100))

I don't know which part that I am doing wrong but my accuracy cannot reach even %1.

Upvotes: 0

Views: 276

Answers (1)

Martin Thoma
Martin Thoma

Reputation: 136389

You have 10000(!) classes. How long do you train? How much training data do you have per class?

Your approach is almost certainly the problem. While you can solve problems "brute force" like this, it is a very bad way to do so. You should first try to detect single digits and then classify each digit with a 10-class classifier.

Upvotes: 2

Related Questions