paulu / deepfeatinterp

Deep Feature Interpolation (CVPR 2017)
GNU General Public License v3.0
278 stars 50 forks source link

Have u tried to transform from low resolution to high resolution image, like super resolution? #8

Open RyanCV opened 6 years ago

RyanCV commented 6 years ago

Hi, thanks for sharing your code. Have u tried to transform from low resolution to high resolution image, like super resolution? Do you have any suggestions if I apply it to SR problem. Thanks.

paulu commented 6 years ago

There is a "Blurry" attribute. Let the negative set be Blurry == True and the positive set be Blurry == False (also matching gender and smile, as usual) then it will sharpen the result.

    elif config.method=='sharpen':
      cP=[(gender,XA[gender]>=0),(smile,XA[smile]>=0),(blurry,True)]
      cQ=[(gender,XA[gender]>=0),(smile,XA[smile]>=0),(blurry,False)]

The input image below was upsampled from 100x100 to 300x300 then I applied DFI to sharpen it.

aaron_blurry aaron_blurry_1520362513_sharpen

MRSHADOW07 commented 1 year ago

1660211737957 i did this

Marcelllo2536 commented 1 month ago

ComfyUI_00008_

paulu commented 1 month ago

@Marcelllo2536 Was this created with DFI?

Marcelllo2536 commented 1 month ago

ComfyUI and Supir.

Marcelllo2536 commented 1 month ago

I was looking for a low quality image to test the Supir model in ComfyUI and I came across this image here on Github, the result I sent still doesn't match what this model is capable of doing.

Marcelllo2536 commented 1 month ago

This one was much more faithful to the original Sem título

shafiur24 commented 1 month ago

Can you tell which model did you use for generating this image?