Xiaojiu-z / Stable-Makeup

Pytorch Implementation of "Stable-Makeup: When Real-World Makeup Transfer Meets Diffusion Model"
Apache License 2.0
130 stars 11 forks source link

about clip image encoder #4

Closed lh-13 closed 6 months ago

lh-13 commented 6 months ago

What should I use instead of clip image encoder? Because clip image encoder will reduce the reference image to 224*224, and some details may be last. so I wonder if there is a feature extrator with a higher resolution ?

Xiaojiu-z commented 6 months ago

What should I use instead of clip image encoder? Because clip image encoder will reduce the reference image to 224*224, and some details may be last. so I wonder if there is a feature extrator with a higher resolution ?

Higher resolution results in higher computational overhead...... I believe one way to enhance the performance of an image encoder is to train its image restoration capabilities. (Maybe you can develop a better feature extractor.)