Of course I did not google the concept of recursive neural networks before I started this experiment and enjoyed the illusion of being very innovative here for a whole day: https://en.wikipedia.org/wiki/Recursive_neural_network …
-
-
Prikaži ovu nit
-
The principle is pretty simple: in a classic residual architecture you chain several residual blocks behind each other (in
#pix2pixHD the default is 9 blocks), what I do in#RecuResGAN is to use a single block, but loop 9 times over it, feeding its output back into its input.Prikaži ovu nit -
The same goes for the down- and up-convolution modules, only here you cannot reduce or increase the amount of filters within a block so you have to compromise a bit how many you use in order not to run out of memory with the accumulated gradients.
Prikaži ovu nit -
So my theory why it seems to be relatively scale invariant is particular because of those recursive up- and down convolutions where a block has to handle all the features at various scales and thus becomes kind of fractal.
Prikaži ovu nit
Kraj razgovora
Novi razgovor -
-
-
Looks like your model is massively overfitting and remembering it’s training data, 3rd image looks like Levitski to me (http://art.niv.ru/images/enc-popular-art/04138.jpg … & https://artsandculture.google.com/asset/portrait-of-countess-ursula-mniszek/IQFUoMwDqzRUWw …)pic.twitter.com/UtMnym4Qgi
-
Did you read the annotation on the image?
- Još 6 drugih odgovora
Novi razgovor -
-
-
How would the model learn higher and lower level features with shared parameters?
-
That is the magic of it. It seems to accumulate all of them in a single layer - of course you cannot extract them later, but since I am not using it for classification that is not an issue for me.
- Još 3 druga odgovora
Novi razgovor -
-
-
Tweet je nedostupan.
-
Yes they are, though these are generated from face markers that are also part of the training data, so these are very easy for the model to make.
Kraj razgovora
-
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.