CompVis / adaptive-style-transfer

source code for the ECCV18 paper A Style-Aware Content Loss for Real-time HD Style Transfer
https://compvis.github.io/adaptive-style-transfer/
GNU General Public License v3.0
723 stars 139 forks source link

Training image volume #25

Open beyondbeneath opened 5 years ago

beyondbeneath commented 5 years ago

How are the results if training on less than 100 GB of images?

If reasonable, is it possible to get an archive of some sampled set (5-10%) of these images from somewhere? The 100 GB of images (which appears to require specific directory structure of the images, not otherwise available in the usual Places365 download) is prohibitively large (both to download & train!) for some setups.

Cheers