2015-11-06, 23:56
Most of the algorithms are closed source. So basically you can only compare the image quality of the resulting image for those. And you can compare the input they take. For example our VDPAU implementation for nvidia can do Motion and Spatial deinterlacing. E.g. compensating the motion and afterwards cleaning up spatial artifacts. It can take 3 past and 3 future frames into account.
Intels' MADI and Intel's MCDI only takes one forward field into account. But the vebox (internal hw since hsw) uses quite advanced algorithms to estimate motion and compensate it. When benchmarking against e.g. Yadif deinterlace from ffmpeg - you won't notice a difference to MCDI at all.
As I have seen so much foobar on this forum I am quite sure that there is a minority of people that understand what I am writing ...
Quality wise:
From left to right - basic -> advanced
Nvidia (VDPAU): Bob, Temporal, Temporal Spatial
AMD (VDPAU): Bob, Temporal (different implementation vs. Nvidia - uses a quite newish algorithm implemented ~2 years ago)
Intel (VAAPI): Bob, MADI, MCDI (works nice on BSW, HSW, BDW, SKL) - only poor Baytrail (BYT) needs to use Bob as it is too slow
AMD (fglrx): Nothing :-) - broken and beyond repair
Software: Bob (Render), Yadif (half), Yadif <- those latter two are called "Deinterlace" in kodi. Yadif has awesome quality but uses a lot CPU - if you use that on chromebox - one core is on 100% - but as MCDI has same image quality, it does not matter.
Intels' MADI and Intel's MCDI only takes one forward field into account. But the vebox (internal hw since hsw) uses quite advanced algorithms to estimate motion and compensate it. When benchmarking against e.g. Yadif deinterlace from ffmpeg - you won't notice a difference to MCDI at all.
As I have seen so much foobar on this forum I am quite sure that there is a minority of people that understand what I am writing ...
Quality wise:
From left to right - basic -> advanced
Nvidia (VDPAU): Bob, Temporal, Temporal Spatial
AMD (VDPAU): Bob, Temporal (different implementation vs. Nvidia - uses a quite newish algorithm implemented ~2 years ago)
Intel (VAAPI): Bob, MADI, MCDI (works nice on BSW, HSW, BDW, SKL) - only poor Baytrail (BYT) needs to use Bob as it is too slow
AMD (fglrx): Nothing :-) - broken and beyond repair
Software: Bob (Render), Yadif (half), Yadif <- those latter two are called "Deinterlace" in kodi. Yadif has awesome quality but uses a lot CPU - if you use that on chromebox - one core is on 100% - but as MCDI has same image quality, it does not matter.