5451
General Software Discussion / Re: Request: Deepdream frontend.
« Last post by mouser on October 01, 2015, 06:29 PM »I just read through the DeepStyle paper -- its similar but different -- it's computing separate representations of an image's content features and "style" (a bit handwavey, but correlations in texture and color), and then generating (by optimizing a loss function) a novel of image that comes close to having the "content" of the one image and the "style" of another.
I haven't read the DeepDream papers, but essentially I believe they are using object recognition data to hallucinate/exaggerate features that are "perceived" (by algorithm but not by human cognition) in an image.
I haven't read the DeepDream papers, but essentially I believe they are using object recognition data to hallucinate/exaggerate features that are "perceived" (by algorithm but not by human cognition) in an image.

Recent Posts
