What are the settings that you use (noise, blur, face refinement)?
I use GP a lot and I have found that the main problem in frame by frame processed videos is with the stars : they are “split” (GigaPixel redraws two sharpened stars instead of one blurred) and they move from frame to frame.
It seems to me that you also have this problem in your video. Did you achieve to reduce or solve it?
I know exactly what you are talking about. I personally have found the best results are on low/low anything more than that just amplifies the mistakes Gigapixel is making. I’ve been thinking about doing a double pass, at low/low and then just a plain scale with no image improvement, and layering the two images to try and reduce the enhanced noise/small details that become distorted.
Thank you very much for sharing your tips!
Did you try doing a double pass, or did you just think about doing it?
On my side, I tried to increase the original frame with a simple bilinear resizing (ia from 720x480 to 960x640) before upscaling it with GP (ia from 960x460 to 1920x1280): that reduces the distorstion of small details, but that reduces also the advantage of using GP…
I have not had a chance to try the double pass & layering of the frames yet, but I will report back when I have had a chance to try it. I was also thinking about trying a single pass on low/low then running a separate pass just as a resize with no filter and using TOOT to average between the two - generating a new frame (as compared to laying two frames on top of one another).
There is a brief discussion about TOOT over in my “Faces” upscale thread.
One thing that I have tried and I know works, but is incredibly cumbersome is “masking.” To me, it seems like GP is great on larger objects that are “up close”/the foreground in the frame but it makes a lot of mistakes on small details in the backgrounds. I’ve successfully made a mask to show just the enhanced detail of the foreground images overlayed over an enlarged background without enhancements - its not great, but its a step in the right direction. It takes a lot of time and the more elements there are in a frame(or scene) the harder it is to pull it off.
What would really be ideal is more control over where GP focuses during its upscale when its attempting to add in details. I’m also giving “Phoenix” a try which could be promising.