How useful is no resizing ?

[MrDarcy]

After cropping a capture with letterbox black borders, I get a perfect 544x320 picture that would need no resize at all (aspect error=0.0 W-zoom=100 H-zoom=100), but I have to keep two black pixels on the top (or bottom) of the picture.

If I crop these two pixels too, I have some aspect ratio error and a 101% H-zoom. To avoid this, I go to the lower resolution, crop a little more, but W-zoom and H-zoom are 96% and the picture is smaller.

So, the question is: how faster is the enconding without need to resize but with a constant 2 black pixels, rather than no black pixels at all but with a zoom to perform ?

Thanks.

[manono]

Hi-

Personally, I will never leave black bars. If it means a small AR error (less than 0.5%), then I can live with that. If it means cutting into the video to get low AR error, or if it means an H-Zoom a bit over 100%, then I can live with that also. Doing those things will allow you to keep it at 544x320. Anything to get rid of all the black.

I'm sure that not resizing means faster encoding, but I don't know how much faster.

[TheWEF]

i don't like black bars at all as well.
how much resizing is slowing down encoding depends on source resolution, output resolution and the resize method (filter type) itself.
test it and let us know your results.

what is resolution and pixelAR of your source? - it's strange that you do not have to resize.

wef.

[MrDarcy]

Good, I was sure that having a zoom > 100% was absolutely to avoid (because of that red colour in the box), but I'm happy to read that 101% can be accepted too.

My source is 544x576 letterbox, captured from DVB-S via SCSI. I cropped 128 black pixels both at the top and at the bottom and I get 544x320, but there still are 2 black pixel at the top.

Input aspect ratio is those I asked about a few days ago here on the Forum. You (TheWEF) suggested 1.448, but 1.459 (PAL 16:9 anamorphic) is good too. So, I set PAR 1:1 in GordianKnot to encode in this almost-anamorphic format, then I will adjust it during playback on my Kiss player and 16:9 TV set (I simply have to select the 'Widescreen' format that is used for really anamorphic DVD's too).

So, I was hoping not to have to resize at all because I think this would be very fast, but those 2 further black lines was a problem.

Now I'm going to do some tests.

Thank you very much.

[MrDarcy]

I did some tests and I think the results are interesting.

I used only 13148 frames taken from the middle of the movie (LOTR)
Conditions: Input resolution 544x576 - Input PAR 1:1 - Bitrate 930 kbit/sec - All crop with even numbers - Bilinear


1) No resize with 2 black pixels left
Crop: left=0 top=128 right=0 bottom=128
Resolution after cropping=544x320
Output resolution=544x320
Error=0 W-Zoom=100 H-Zoom=100
Fps: 1st pass=19.568 2nd pass=29.745


2) Enlarge to remove black pixels
Crop: left=2 top=130 right=2 bottom=128
Resolution after cropping=540x318
Output resolution=544x320
Error=-0.1 W-Zoom=101 H-Zoom=101
Fps: 1st pass=17.141 2nd pass=25.117

Now I reduced the output resolution

3) Minimum reduce without any black pixels
Crop: left=4 top=134 right=4 bottom=134
Resolution after cropping=536x308
Output resolution=528x304
Error=+0.2 W-Zoom=99 H-Zoom=99
Fps: 1st pass=17.895 2nd pass=25.521


4) More cropping to avoid resizing
Crop: left=8 top=136 right=8 bottom=136
Resolution after cropping=528x304
Output resolution=528x304
Error=0 W-Zoom=100 H-Zoom=100
Fps: 1st pass=20.912 2nd pass=32.491


At the playback on the PC without zoom I see no differences, tomorrow I will try to burn the AVIs and play them on the Kiss+TV.

Now I go to sleep.... I leave comments to expert guys....

Bye !

[N_F]

I'm surprised at the difference of the speed between the first and second pass. I get ~20% better speed on second pass, you seem to get 50% :confused:

[TheWEF]

Originally posted by MrDarcy
You (TheWEF) suggested 1.448, but 1.459 (PAL 16:9 anamorphic) is good too. So, I set PAR 1:1 in GordianKnot to encode in this almost-anamorphic format

i see, so "aspect error=0.0" was not really true, because you already entered a "wrong" par (at least a par, nobody can confirm atm, because of the strange source resolution).

wef.

[MrDarcy]

Originally posted by TheWEF
i see, so "aspect error=0.0" was not really true, because you already entered a "wrong" par (at least a par, nobody can confirm atm, because of the strange source resolution).

wef.

Yes, in my case "aspect error=0" means: no further aspect error caused by resizing.

Let's suppose that I know the correct PAR for that kind of resolution (and I'm going to investigate about it), how can I calculate the custom PAR value to enter in GKnot to encode in a perfect anamorphic ratio, so that setting my TV to 'widescreen' I'll get the correct picture?

---

To N_F: I don't know, these values are reported by GKnot itself.

Maybe it's due to the simple Bilinear filter I used for this test.
Is the resizing filter applied during the second pass, isn't it ?
(I'm still very newbie, I know.....)

[TheWEF]

Originally posted by MrDarcy
Y...how can I calculate the custom PAR value to enter in GKnot to encode in a perfect anamorphic ratio...

PAR_non_ananamorphic * (4/3) = PAR_anamorhic

wef.