In "4.1 Learning multi-layer deconvolutional filters" section, the last paragraph says that "Since our model is generative, we can sample from it. In Fig. 3 we show samples from the two different models from each level projected down into pixel space. The samples were drawn using the relative firing frequencies of each feature from the training set."

Fig.3 and the explanition

I don't know how the pictures in Fig.3 are generative. Since that the filters has been learned and feature maps in every layer can be infered, for example, in terms of fruit samples, in "Layer 1", is the the first layer's feature map ? i feel that not true...seems like the sample are low-level...paper says "... from each level projected into pixel space", these words are short and confuse me.

somebody could explain that for me? thank you very much!


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.