MrMC

cosmoxl wrote: 09 Feb 2018, 02:06 As I've grown more experienced with using yadif 2x on the ATV4k I've come to the conclusion that the algorithm isn't looking at the whole screen to decide cadence but only (probably) the center. Whenever the image in the center of the screen is still, yadif 2x is thrown off and skipping or dropping happens. As soon as the center starts moving it locks on to the cadence again.

My requests is that if there are deep settings in the yadif2x code that control what part of the image is inspected, that that area of inspection be increased to include the whole image or a bigger part of the image. Even better is to possibly reveal that setting to the user.

Thanks!

davilla wrote: 09 Feb 2018, 22:27

cosmoxl wrote: 09 Feb 2018, 02:06 As I've grown more experienced with using yadif 2x on the ATV4k I've come to the conclusion that the algorithm isn't looking at the whole screen to decide cadence but only (probably) the center. Whenever the image in the center of the screen is still, yadif 2x is thrown off and skipping or dropping happens. As soon as the center starts moving it locks on to the cadence again.

My requests is that if there are deep settings in the yadif2x code that control what part of the image is inspected, that that area of inspection be increased to include the whole image or a bigger part of the image. Even better is to possibly reveal that setting to the user.

Thanks!

Investigate yadif for FFMpeg, I'm sure they have lots of diddle factors

noggin wrote: 30 Apr 2018, 11:45 Sorry to necropost - but W3FDIF (aka Weston 3 Field Deinterlacing) is a good option to consider.

It's non-adaptive (so doesn't try to detect motion), but instead it is based on a clever Vertical/Temporal filter. It was patented by the BBC, and licensed by a number of broadcast manufacturers who needed deinterlacers in their digital video equipment - say for high-end picture manipulating digital video effects devices, or frame rate conversion. Once the patent had expired someone from the BBC ported it into ffmbc (a broadcast fork of ffmpeg) and it is now also in ffmpeg.

Because it's non-adaptive it seldom 'gets things wrong', and it's also very suitable for GPU processing AIUI (not relevant to the ATV)

If the ATV4K has enough CPU power, it may be worth seeing if it can do a W3FDIF rather than a YADIF 2x?

Whenever I do any off-line deinterlacing in ffmpeg I use the video filter "w3fdif=complex:all" these days.

noggin wrote: 30 Apr 2018, 11:45 Sorry to necropost - but W3FDIF (aka Weston 3 Field Deinterlacing) is a good option to consider.

It's non-adaptive (so doesn't try to detect motion), but instead it is based on a clever Vertical/Temporal filter. It was patented by the BBC, and licensed by a number of broadcast manufacturers who needed deinterlacers in their digital video equipment - say for high-end picture manipulating digital video effects devices, or frame rate conversion. Once the patent had expired someone from the BBC ported it into ffmbc (a broadcast fork of ffmpeg) and it is now also in ffmpeg.

Because it's non-adaptive it seldom 'gets things wrong', and it's also very suitable for GPU processing AIUI (not relevant to the ATV)

If the ATV4K has enough CPU power, it may be worth seeing if it can do a W3FDIF rather than a YADIF 2x?

Whenever I do any off-line deinterlacing in ffmpeg I use the video filter "w3fdif=complex:all" these days.