Tensorflow 之finetune微调模型方法&&不同层上设置不同的学习率

• 在不同层上设置不同的学习率，fine-tuning

• https://github.com/dgurkaynak/tensorflow-cnn-finetune

ConvNets:

• AlexNet
• VGGNet
• ResNet

AlexNet finetune:

- 自己搭建的网络，加载初始化模型：

def load_with_skip(data_path, session, skip_layer):
    data_dict = np.load(data_path).item()
    for key in data_dict:
        if key not in skip_layer:
            with tf.variable_scope(key, reuse=True):
                for subkey, data in zip(('weights', 'biases'), data_dict[key]):
                    session.run(tf.get_variable(subkey).assign(data))

print('Load pre-trained model: {}'.format(weight_file))
load_with_skip(weight_file, sess, ['fc8']) # Skip weights from fc8

• https://github.com/joelthchao/tensorflow-finetune-flickr-style
• https://github.com/kratzert/finetune_alexnet_with_tensorflow

VGG模型finetune:

- 自定义网络；加载参数，很详细教程：

• Tensorflow学习笔记：CNN篇（9）——Finetuning，复用在ImageNet已训练好的VGGNet进行图像识别
• Tensorflow学习笔记：CNN篇（10）——Finetuning，猫狗大战，VGGNet的重新针对训练
• https://www.cs.toronto.edu/~frossard/post/vgg16/

- 另外：这是个基于tensorflow-vgg16和Caffe to TensorFlow的VGG16和VGG19的一个TensorFlow的实现。

• https://github.com/leihe001/tensorflow-vgg

ResNet的slim模型finetune:

• https://github.com/Shirhe-Lyh/finetune_classification

• 使用tensorflow 的slim模块fine-tune resnet/densenet/inception网络，解决batchnorm问题