MCBob yes, is from dideč, MVBob is from scharfis_brain. Re-touchč! :wink2:
TDeint in default mode is also not lossless IIRC Edit: Quote:
P.S.: a good deinterlacer is not lossless |
To clarify: by default, Yadif is a single-rate deinterlacer. With mode=1 it's a bobber (double-rate deinterlacer). You listed "Yadiff without mode=1", well without double-rate output of course it does much worse than altering the fields, it throws out half of them.
|
int mode = 0
0 : single frame rate, temporal and spatial interlacing check (default). 1 : double frame rate, temporal and spatial interlacing check. 2 : single frame rate, skips spatial interlacing check. 3 : double frame rate, skips spatial interlacing check. |
The sad part is that I actually read that section of the Wiki before posting. :o I'll take your word for it that mode=3 is lossy; never used it.
|
I did several experiment to find a lossless deinterlacer for my restoration flow, but it was loooong time ago, so do no trust me too much :wink2:
|
Quote:
|
Quote:
|
Few posts above, I didn't post the samples publicly, just for my own testing, I upload TV commercials now.
|
Quote:
|
I did, as I said few posts above, I think post #12, It's an enfant channel with much room to grow, though I'm not willing to make any money out of it, just sharing the memories.
|
I guess that the latreche34 YouTube channel isn't anything to do with you then?
|
Quote:
|
2 Attachment(s)
Quote:
Quote:
https://www.youtube.com/watch?v=SiQj3k4orIo&t=62s which has been uploaded after a x264 compression with crf=17. I uploaded also a small sequence without x264 compression, using the HuffYUV file. I then downloaded the 2 videos elaborated by youtube with JDownloader 2 and compared them. video comparison: Attachment 14135 image comparison: Youtube after processing: mp4 versus huffYUV https://imgsli.com/NzA5MDQ I see no significat difference, which I suspect to be not good, because it means that the youtube compression (applied on both) may hyde the small quality difference of the lossless file However, when I also compare the files before upload (and then before youtube compression) I see a very very small difference in term of quality (i reinvented the whell, low crf x264 compression is almost transparent) video comparison: Attachment 14134 image comparison: Youtube source: mp4 versus huffYUV https://imgsli.com/NzA5MDM For my videos there is not significant difference after upload if a crf=17 x264 compression is performed before upload, so, while for small duration video I will still upload huffYUV version, for long video I will compress to x264 with crf=17 before upload. |
Just a side note, When resizing to 1080 there is no reason to keep the black borders around the frame, The resizing of lines and pixels will not be an integer ratio, the mathemathical approximation will have to be done anyway so may as well remove the boarders and keep a clean frame.
|
Good hint, the problem is that I have a different "active" image in each capture, so, in the fear of introducing a distortion, I add the black borders removed during the filtering and then resize to 1440x1080, by default.
Specially for letterbox video the final result is horrible, I will try to improve it. |
I set custom visual cropping for each tape, Commercials are tricky, they don't have the same active area so I find the sweet spot for all, AVSPmod allows me to scroll through the time line and see each segment so the final crop works for most of the video, once satisfied I hit apply, It also allows me to see the de-interlacing results and resizing instantly, pretty neat.
If you look at my videos you'll still see some odd ball commercials with black bars left over but 90% of the video looks fine. The TV overscan days took away more than what we crop now in digital capture, almost like 20 pixels on each side but we never noticed. |
I do the same, and just as example, this is a set of some of the "active area" I found on my 720x576 captures:
# width x height # 698x544 # 688x560 # 690x560 # 690x562 # 688x564 # 704x552 # 688x570 # 694x566 # 694x570 :eek: :eek: Assuming that the 4:3 PAL SD aspect ratio is based on the 720x576 non-square pixel capture, or better on the 704x576 non-square pixel area, if I crop every video to its active area and then blindly resize to 1440x1080, the "proportion" of one will not be the same of another, and not by a negligible amount :paranoid: That's why I keep the original black borders, but may be I am just over-processing... |
Unfortunatly it is based on 704x576 for PAL/SECAM and 704x480 for NTSC (refer to the circle test I made over at VH), But for cropping you can drift away from that ratio by up to 5 pixels vertically or horizontally before it can be noticeable, and honestly most people don't even notice 720x576 and 720x480 directly formatted into 4:3. Some are even okay with 4:3 being stretched to 16:9, even some clueless technicians did it at the proffesional level in TV stations back in the day.
|
1 Attachment(s)
Quote:
Quote:
Quote:
Edit: even if it not a crime like 4:3 being stretched to 16:9, but rather a "mutilation", here a sample of commercial vhs and dvb-s dump of the same movie through the years: Attachment 14142 |
Quote:
Quote:
|
Site design, images and content © 2002-2024 The Digital FAQ, www.digitalFAQ.com
Forum Software by vBulletin · Copyright © 2024 Jelsoft Enterprises Ltd.