Why did you use VGG16 instead of Resnet?

[scut-salmon]

Hi, I found that the model just overfits in just some epochs, is this because the backbone is VGG instead of resnet?

[hcnhatnam]

I have same problem.

[clement10601]

you can change the backbone

[Zhang-O]

and how to change the backbone.

[scut-salmon]

@Zhang-O @hcnhatnam It's ok to change to resnet50 by two lines, but it seems no improvement. vgg16 = ResNet50(input_tensor=self.input_img, weights='imagenet', include_top=False) self.f = [vgg16.get_layer('activation_49').output, vgg16.get_layer('activation_40').output, vgg16.get_layer('activation_22').output, vgg16.get_layer('activation_10').output]

[Zhang-O]

3ks,could you add me to qq friends? my account is 691259192

[Zhang-O]

I am new to advanced East .i think I will have lots questions when study east and advanced east

[Zhang-O]

????

[Zhang-O]

I have flowed you

[ccsone]

@Zhang-O @hcnhatnam It's ok to change to resnet50 by two lines, but it seems no improvement. vgg16 = ResNet50(input_tensor=self.input_img, weights='imagenet', include_top=False) self.f = [vgg16.get_layer('activation_49').output, vgg16.get_layer('activation_40').output, vgg16.get_layer('activation_22').output, vgg16.get_layer('activation_10').output]

@scut-salmon HELLO i changed resnet50 based your method， it occurs out dimension error