Ai Dreams Forum
Artificial Intelligence => General AI Discussion => Topic started by: LOCKSUIT on December 18, 2018, 08:00:48 pm
-
AI GAN TAKES A WHOLE NOTHER LEAP & U CAN TRY IT O M G GOTTA SEE THIS!!!!!!!
https://www.youtube.com/watch?v=ZKQp28OqwNQ
-
Here is a link. This was a lot of fun! Sort of brings out the "Mad scientist" in us all.
https://colab.research.google.com/github/tensorflow/hub/blob/master/examples/colab/biggan_generation_with_tf_hub.ipynb#scrollTo=dSAyfDfnVugs (https://colab.research.google.com/github/tensorflow/hub/blob/master/examples/colab/biggan_generation_with_tf_hub.ipynb#scrollTo=dSAyfDfnVugs)
-
I liked it too - fast, big, adjustable, interpolation, realistic........however the whole time I was let down at one major craving - I wanted to use my own image and interpolate it or upsize it, and way more than just 512x512. No fun...
It seems 512x512 chooses a path to that module. Can't increase it.
-
@Hinton
@Korrelan
Food for thought:
BigGANs said they make it get good at 1 view of them looking right at you, to help with not knowing how many legs to give spiders. Hinton is working on making cortical columns have different views of an object. Cortical columns may achieve redundancy (so a memory no easily dies if 1 neuron dies) by being invariants of different angels. ?