search

machrisaa / tensorflow-vgg

VGG19 and VGG16 on Tensorflow
2.21k stars 1.08k forks source link

Train on my own data... #66

Open JayjieL opened 5 years ago

JayjieL commented 5 years ago

Hello,when i use vgg19_trainable to train my own data from scratch, i got the problem that the loss and acuuracy on test data keep the same, it seems that the network is not working.Have you ever met such a problem?Thank you in advance for your contribution and hope to hear from you soon.　

SdustZhangzhen commented 5 years ago

@JayjieL Hello, I also want to train my own data about Remote Sensing image. But I have no ideal about how to make the label. Can you give some suggests? Thanks. 你好，我也想用VGG19训练自己的样本数据，但是不知道怎么制作标签和训练数据，您有什么好的建议吗？谢谢！

JayjieL commented 5 years ago

@JayjieL Hello, I also want to train my own data about Remote Sensing image. But I have no ideal about how to make the label. Can you give some suggests? Thanks. 你好，我也想用VGG19训练自己的样本数据，但是不知道怎么制作标签和训练数据，您有什么好的建议吗？谢谢！

File vgg19_trainable.py already provide the base network structure. So, just modify the input layer to your image size(or you can resize your pictures), and change the output dimension. As for making labels, you can manually mark them(or try to use labeled data)

Ruolingdeng commented 5 years ago

Do you know how to implement the "use this to build the VGG object", Please?

usmanxia commented 5 years ago

@JayjieL were you able to train the model? I am exactly facing the same problem you mentioned that the loss and accuracy do not change. Any help is really appreciated.

JayjieL commented 5 years ago

@usmanxia It seems that there is an integrated interface under the Keras framework, which can be called directly.（I have not used this network anymore）