Can't reproduce the performance on real_example charlie.png

assafshocher / ZSSR

"Zero-Shot" Super-Resolution using Deep Internal Learning

Other

401 stars 102 forks source link

Can't reproduce the performance on real_example charlie.png #8

Open back2yes opened 5 years ago

back2yes commented 5 years ago

Hi, I ran the code run_ZSSR_single_input.py on charlie.png but the results seemed to be about the same with EDSR, this is what I ran:

python run_ZSSR_single_input.py real_example/charlie.png 0 0 0 X2_GRADUAL_IDEAL_CONF output_real

I am wondering if I misunderstood the code?

assafshocher commented 5 years ago

Hi, You ran a real image (non-ideal) with ideal conf. To run on this particular image, just run: python run_ZSSR.py X2_REAL_CONF

back2yes commented 5 years ago

Thank you for your help, I tried X2_REAL_CONF but the result seemed to be the same

This is what I get https://pbs.twimg.com/media/DsKk6i7VsAAL8XW.jpg:large, could you help to judge if this result is correct?

assafshocher commented 5 years ago

Hi, please make sure the kernel .mat file exists and that you don't get a message that no kernel is used when running.

CurryYuan commented 4 years ago

Same problem here.

assafshocher commented 4 years ago

@back2yes, @CurryYuan First, which Python version do you use? ZSSR is made for py2 (unfortunately). If you run it with py3, and modified, you may have problems with some divisions ( / ) in the resizing. This is slovable and stable py3 versions exist. let me know. Second, have you seen the kernel loaded during the ZSSR run? When it is loaded, this line prints the kernel file path. However, if no kernel supplied then this line prints "no kernel loaded".