search

TencentARC / T2I-Adapter

T2I-Adapter
3.47k stars 208 forks source link

StyleAdapter training datasets? #33

Open CiaoHe opened 1 year ago

CiaoHe commented 1 year ago

Just curious about how to construct the style pairs? any mature public datasets available?

ziye3001 commented 1 year ago

well, I have the same question. Style transfer is not like the other tasks which you easily get the ground-truth data. Besides, it 's a bit wired to simply take the addition of the features from the encoder and the ones from the T2I-Adapters, because the content image and the style image are, in general, not spatially related.

MC-E commented 1 year ago

We did not use any training pairs, only the style image. The details will be released soon.

asheroin commented 1 year ago

We did not use any training pairs, only the style image. The details will be released soon.

Where would it be released? Just a simple update in README or, if necessary, posted to another repo?