I want to upscale my HVD files which were shot on a SONY FX1 fifiteen years ago.
The source file is a 1440x1080i50 HDV file. The destination should be a sharper and upscaled 3840x2160 version.
Encoding with DDV-2 (50p) results in a flickering background. You can see the leeve borders switching from dark to a kind of halo each frame.
Encoding with DTD-3 (25p) is too jerky and the borders of the leeves are mostly dark. Nevertheless, 25p will not get good results.
The source file was then deinterlaced with ffmpeg to 50p (HEVC) using the bwdif method (field, top, all frames). This file does not show any flickering if you check the single frames step by step
The 50p file was now encoded with ghq-5 and ahq-11 (i also tried ahq-10 and Theia from which none create better results)
Both results do seem not as sharp and good as the DDV-2 version despite of the flickering. You can see this if you look at the hat, the beard and also the big black ring at the chest of the man. DDV-2 does seem to create significantly more details than any other method.
So i simply tried to upscale the 50p file using DDV-2, which in fact creates the result i want, but it of course doubles the frames again resulting in a 100fps file, which then needs to be re-encoded again to get 50p. This is a step too much.
I then took the DDV-2 model data and changed the configuration (JSON) to progressive using a new type called DDP-2 also duplicating the DDV files and change their names to DDP. Now i have a new model available which can upscale progressive sources but using the DDV-2 algorythms.
This is the custom DDP-2 image:
This is GHQ-5 from the same 50p source file:
This is AHQ-11 from the same 50p source file:
You need to zoom in quite a bit to see the difference, but it’s very noticeable, that DDP-2 based on DDV-2 is sharper and creates more details.
Nevertheless, the build-in deinterlacer does not really work for this type of footage due to the resulting flickering, which even gets worse if you slow down 50p to 24p.!
