张量流softmax如何添加未知类?

时间:2018-12-27 09:09:46

标签: python tensorflow softmax

我使用Tensorflow建立了ocr分类系统。

以下是图形:

def build_graph(top_k):
    # with tf.device('/cpu:0'):
    keep_prob = tf.placeholder(dtype=tf.float32, shape=[], name='keep_prob')
    images = tf.placeholder(dtype=tf.float32, shape=[None, 64, 64, 1], name='image_batch')
    labels = tf.placeholder(dtype=tf.int64, shape=[None], name='label_batch')

    conv_1 = slim.conv2d(images, 64, [3, 3], 1, padding='SAME', scope='conv1')
    max_pool_1 = slim.max_pool2d(conv_1, [2, 2], [2, 2], padding='SAME')
    conv_2 = slim.conv2d(max_pool_1, 128, [3, 3], padding='SAME', scope='conv2')
    max_pool_2 = slim.max_pool2d(conv_2, [2, 2], [2, 2], padding='SAME')
    conv_3 = slim.conv2d(max_pool_2, 256, [3, 3], padding='SAME', scope='conv3')
    max_pool_3 = slim.max_pool2d(conv_3, [2, 2], [2, 2], padding='SAME')

    flatten = slim.flatten(max_pool_3)
    fc1 = slim.fully_connected(slim.dropout(flatten, keep_prob), 1024, activation_fn=tf.nn.tanh, scope='fc1')
    logits = slim.fully_connected(slim.dropout(fc1, keep_prob), FLAGS.charset_size, activation_fn=None, scope='fc2')
    loss = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=logits, labels=labels))
    accuracy = tf.reduce_mean(tf.cast(tf.equal(tf.argmax(logits, 1), labels), tf.float32))

    global_step = tf.get_variable("step", [], initializer=tf.constant_initializer(0.0), trainable=False)
    rate = tf.train.exponential_decay(2e-4, global_step, decay_steps=2000, decay_rate=0.97, staircase=True)
    train_op = tf.train.AdamOptimizer(learning_rate=rate).minimize(loss, global_step=global_step)
    probabilities = tf.nn.softmax(logits)

    tf.summary.scalar('loss', loss)
    tf.summary.scalar('accuracy', accuracy)
    merged_summary_op = tf.summary.merge_all()
    predicted_val_top_k, predicted_index_top_k = tf.nn.top_k(probabilities, k=top_k, name="predicted_top_k")
    accuracy_in_top_k = tf.reduce_mean(tf.cast(tf.nn.in_top_k(probabilities, labels, top_k), tf.float32))

    return {'images': images,
            'labels': labels,
            'keep_prob': keep_prob,
            'top_k': top_k,
            'global_step': global_step,
            'train_op': train_op,
            'loss': loss,
            'accuracy': accuracy,
            'accuracy_top_k': accuracy_in_top_k,
            'merged_summary_op': merged_summary_op,
            'predicted_distribution': probabilities,
            'predicted_index_top_k': predicted_index_top_k,
            'predicted_val_top_k': predicted_val_top_k}

这是推理功能:

def inference(images, pbfile="pb/ocr.pb"):
    print('inference')
    start = time.time()
    predicted_val_top_k = graph.get_tensor_by_name('ocr/predicted_top_k:0')
    predicted_index_top_k = graph.get_tensor_by_name('ocr/predicted_top_k:1')
    tensor_image = graph.get_tensor_by_name('ocr/image_batch:0')
    keep_prob = graph.get_tensor_by_name('ocr/keep_prob:0')
    probabilities = graph.get_tensor_by_name('ocr/Softmax:0')
    logits = graph.get_tensor_by_name('ocr/fc2/BiasAdd:0')
    end = time.time()
    print('takes %s second to get tensor' % (start - end))
    result = []
    for image in images:
        temp_image = Image.open(image).convert('L')
        temp_image = temp_image.resize((IMAGE_SIZE, IMAGE_SIZE), Image.ANTIALIAS)
        temp_image = np.asarray(temp_image) / 255.0
        temp_image = temp_image.reshape([-1, 64, 64, 1])
        start = time.time()
        logit, prob, predict_val, predict_index = sess.run([logits, probabilities, predicted_val_top_k, predicted_index_top_k],
                                              feed_dict={tensor_image: temp_image, keep_prob: 1.0})
        end = time.time()
        print('takes %s second to run tensor' % (start - end))
        result.append({'image': image, 'val': predict_val, 'index': predict_index})
        document_dict = {
                            0: 'V',
                            1: 'X',
                            2: 'U'
        }
        image_name = image.split('/')[-1]
    return result

我们现在只有三个类,即'V','X','U',如果我们要检测的目标属于这三种类型,则一切正常。

但是,当我们检测到目标不属于该类型的候选对象时,就会出现问题。我们现在将“ A”作为推论,问题是,我们也为“ A”获得了“ X”类,这显然是错误的。

然后,我想通过设置分数阈值来区分其他人。

我们知道tf.nn.softmax返回类似分数的信息,并且当我调试时,我发现目标'A'的类'X'的分数(预测函数中的predict_val)几乎为1(实际上为0.9999)。 )。

然后,在深入到softmax之后,我认为这是合理的:

  softmax = tf.exp(logits) / tf.reduce_sum(tf.exp(logits), axis)

它仅对当前类登录进行操作。

那么,有没有办法为涉及所有其他目标的softmax添加未知类?

Env:Python3.6.5; Tensorflow 1.8.0

谢谢。

卫斯理

1 个答案:

答案 0 :(得分:0)

我将添加一个附加的输出类Unknown,因为这将使您的模型能够学习对训练集中的项目进行分类,同时还具有一个输出以转储与目标之一不完全匹配的所有项目。

对您的代码进行猜测,因为您没有提供有效的示例,所以我认为以下更改可以解决问题:

logits = slim.fully_connected(slim.dropout(fc1, keep_prob), FLAGS.charset_size+1, activation_fn=None, scope='fc2')

现在您的输出将具有4个概率,因此document_dict如下所示:

document_dict = {
                            0: 'V',
                            1: 'X',
                            2: 'U',
                            3: 'Unknown'
        }

您需要了解它的训练方式,但是我希望您现在面对未知输入时,所关心的值(V,X,U)的分数会低得多。