search

zhenngbolun / S-Net

S-Net: A Scalable Convolutional Neural Network for JPEG Compression Artifact Reduction
24 stars 5 forks source link

Is it possible to run a atrifact reduction without gt? #4

Open FabianBartels opened 5 years ago

FabianBartels commented 5 years ago

I´ve run some tests but it looks like i need the HR version of the Image to reduce the artifacts in a LR Version with QF40.

You´ve testet images downloaded (WIN143) from the internet to test the generalization of your model. I guess the wallpapers in WIN143 did not contain any artifacts and you´ve produced artificial artifacts due to compression? Am i right?

Because Im currentry try to reduce artifacts with only one image that contains the artifacts (image with artifacts as gt). Is this possible`?

Thanks again for providing :)

Greets

zhenngbolun commented 5 years ago

For WIN143, we provided the original 1920x1080 HR images. For the resolution of 1920x1080 is too high to put the image in memory when processing. So we used bicubic interpolation (in Matlab) to downsample them to 960x540, and all the evaluations are conducted on downsampled images.