Is ACO (Ant Colony Optimization) an deterministic optimization that can be used?
This is still a non-convex problem right? Or can it be posed as a convex optimization (and then can use interior point methods?)
We have keras+tf extracting the VGG features. These are used by scipy for bfgs. Does the bfgs also run on the GPU?
Where do the semi-regular geometric patterns in the generated image come from?
Follow up to previous? How would batching work for this vgg + scipy scenario?
Has anyone tried something like this by averaging (or in some way combining) the activations of multiple bird images, such that it creates some kind of novel or āprototypicalā bird?
where to get the file āfnames.pklā
Just to make sure I understand this ā in this example we started with a random image. But if we started with the ACTUAL image as the initial condition, we would get the original image back, right? Cause that would perfectly optimize with zero loss right? Iām sure youāll get to it ā but Iām interested to find out where we initialize for the artistic styling problemā¦
Would it be useful to use a tool like Quiver to figure out which Vgg layer to use for this?
Gut-feeling: It looks to me like JPEG compression artifacts. Apparently, VGG is pulling these out since it looks for edges and other features.
Has anyone since the publication of this paper used any other loss functions for f(style) that seem to achieve visually similar results?
If you import the VGG model that is built into keras, do you still have to re-order the channels, etc.?
How do we use the output of the different layers (more than one from VGG) for style loss?
I believe so. keras.applications.imagenet_utils has vgg_preprocess that does this.
are there any kind of homework or other recommendation of deepen the knowledge of the first lecture?
Is there any slack channel for part-2?
No there isnāt a slack channel for part 2. Jeremy said the discussions would be on the forums this time.
Will there be a way to watch again, I missed about 15 minutes and is giving me an unknown error and says canāt watch on a different site, click to watch on youtube (Iām on youtube) when clicking it gives another error has occured.
You can watch livestream again on Jeremyās channel:
Yes, you do