In the reference of vgg16, the first method of feature extraction is to use the front layer of vgg16 to predict, and then spread out the results as input, and then connect its own layer to train. The second method is to directly connect vgg16 to two layers for training. What is the difference between the two methods? The parameters of vgg16 are also frozen.
Hello, the author. I'd like to ask
In the reference of vgg16, the first method of feature extraction is to use the front layer of vgg16 to predict, and then spread out the results as input, and then connect its own layer to train. The second method is to directly connect vgg16 to two layers for training. What is the difference between the two methods? The parameters of vgg16 are also frozen.
Forget to solve doubts