tensorflow 训练fruit3602

最新推荐文章于 2021-06-21 15:30:03 发布

原创最新推荐文章于 2021-06-21 15:30:03 发布 · 445 阅读

2 ·

CC 4.0 BY-SA版权

深度学习专栏收录该内容

12 篇文章

订阅专栏

本文介绍使用VGG16卷积神经网络进行图像分类任务的过程，包括模型定义、学习率设置及训练流程，并展示了训练效果。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

数据预处理完整后，开始训练模型：这里偷了一个懒用了slim模型：

def vgg_model(inputs,
           num_classes=classes,
           is_training=True,
           dropout_keep_prob=0.5,
           spatial_squeeze=True,
           scope='vgg_16',
           fc_conv_padding='VALID',
           global_pool=False):
  with tf.variable_scope(scope, 'vgg_models', [inputs]) as sc:
    end_points_collection = sc.original_name_scope + '_end_points'
    # Collect outputs for conv2d, fully_connected and max_pool2d.
    with slim.arg_scope([slim.conv2d, slim.fully_connected, slim.max_pool2d],
                        outputs_collections=end_points_collection):
      net = slim.repeat(inputs, 2, slim.conv2d, 64, [3, 3], scope='conv1')
      net = slim.max_pool2d(net, [2, 2], scope='pool1')#50
      net = slim.repeat(net, 2, slim.conv2d, 128, [3, 3], scope='conv2')
      net = slim.max_pool2d(net, [2, 2], scope='pool2')#25
      net = slim.repeat(net, 2, slim.conv2d, 256, [3, 3], scope='conv3')
      net = slim.max_pool2d(net, [2, 2], scope='pool3')#13
      net = slim.repeat(net, 2, slim.conv2d, 512, [3, 3], scope='conv4')
      net = slim.max_pool2d(net, [2, 2], scope='pool4')#7
      net = slim.repeat(net, 3, slim.conv2d, 512, [3, 3], scope='conv5')
      net = slim.max_pool2d(net, [2, 2], scope='pool5')#4
      net = slim.conv2d(net, 2048, [3, 3], padding=fc_conv_padding, scope='fc6')
      net = slim.dropout(net, dropout_keep_prob, is_training=is_training,
                         scope='dropout6')
      net = slim.conv2d(net, 2048, [1, 1], scope='fc7')
      # Convert end_points_collection into a end_point dict.
      end_points = slim.utils.convert_collection_to_dict(end_points_collection)
      if global_pool:
        net = tf.reduce_mean(net, [1, 2], keep_dims=True, name='global_pool')
        end_points['global_pool'] = net
      if num_classes:
        net = slim.dropout(net, dropout_keep_prob, is_training=is_training,
                           scope='dropout7')
        net = slim.conv2d(net, num_classes, [1, 1],
                          activation_fn=None,
                          normalizer_fn=None,
                          scope='fc8')
        if spatial_squeeze:
          net = tf.squeeze(net, [1, 2], name='fc8/squeezed')
        end_points[sc.name + '/fc8'] = net
      return net,_

定义model的learning-rate：

cross_entropy = tf.reduce_mean (tf.nn.softmax_cross_entropy_with_logits_v2( labels=y_, logits=logits))
train_step = tf.train.AdamOptimizer(0.001).minimize(cross_entropy)
accuracy = tf.reduce_mean(tf.cast(tf.equal(tf.argmax(logits, 1), tf.argmax(y_, 1)), tf.float32))

训练开始：

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    coord = tf.train.Coordinator()
    threads = tf.train.start_queue_runners(sess=sess, coord=coord)
    base = []
    while True and not coord.should_stop():
        train_batch , train_label = sess.run([train_x,train_y])
        #print(train_label)
        train_label = data.ChangeOneHot(train_label)
        _,loss = sess.run([train_step, cross_entropy],feed_dict = {x:train_batch,y_:train_label,keep2:0.5,train:True,keep1:0.5})
        if index % 200 == 0:
            test_batch , test_label = sess.run([test_x,test_y])
            test_label = data.ChangeOneHot(test_label)
            #print(test_label)
            acc = accuracy.eval({x:test_batch, y_:test_label,keep2:1.0,train:True,keep1:1.0 })
            print(get_now_time(),'step:',index, "acc:", acc, "  loss:", loss)
            getlogs(get_now_time()+' step :'+str(index)+' acc:'+str(acc)+' loss: '+str(loss))
            if acc>0.95 and index >1000:
                #print(get_now_time(),index, "acc:", acc, "  loss:", loss)
                saver.save(sess, save_model + '/train.ckpt', global_step = index)
            if index ==2000:
                #print(get_now_time(),index, "acc:", acc, "  loss:", loss)
                base.append(acc)
                saver.save(sess, save_model + '/train1.ckpt', global_step = index)
            if index > 2000 and base:
                before = base.pop()
                base.append(before)########################
                if acc > before :
                    base.append(acc)
                    saver.save(sess, save_model + '/train1.ckpt', global_step = index)
        index += 1
        if index == 50000:
            saver.save(sess, save_model + '/train.ckpt', global_step = index)
            break
    coord.request_stop()
    coord.join(threads=threads)

训练的截图显示：