Hacker News new | ask | show | jobs
by MayeulC 2644 days ago
There seems to be plenty of less-than-pleasant outputs from this net. I had a couple of strange ones when I landed: [1][2]. If those were generated with GAN, the adversarial network could really use another training pass with examples like this (and if GANs are not used, there could be a simple network filtering the first net's output).

[1]: https://havetheyfaked.me/d/539e20e9-560c-4a8d-a2ba-ca89e2c9a...

[2]: https://havetheyfaked.me/d/86526a1d-bbe7-4956-8bec-a7e64e082...