VAF-5D2 Optical Anti-Aliasing Filter Test on 5D2

Started by Volumetrik, April 23, 2020, 07:00:40 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Volumetrik

I wanted to test out the effect of the VAF filter on the video on the 5D Mark II. As you all know, the 5D2 suffers from aliasing in the full frame mode. On close up subjects it can be fine and not show at all, but on repeating patterns, clothing and fine contrasting detail it can completely ruin the video.

The hypothesis :

I think the VAF filter will eliminate the aliasing problem for full frame video. I also think that it may reduce sharpness.
Additionally, I think using a VAF filter AND crop mode will yield reduced sharpness, but I don't know to what degree.

Variables :

Canon 5D Mark II 2.1.2 w/94k actuations
Canon 35mm F/2 IS @ F/8
Canon EF-S 10-18 STM @ F/8
VAF 5D2 AA Filter
SanDisk ExtremePro 32Gb UDMA 7 160MB/s CF card

magiclantern-crop_rec-3k_Updated_Center_4.20pm-5D2-eXperimental.2019Nov14.5D2212 from Reddeercity
100 ISO
FF mode : 1866 x 1248 1.00x Crop @ 23.976 fps 1/50s
Crop mode : 2784 x 1080 2.01x Crop @ 23.98 fps 1/50s

The setup :

Using a tripod, I shot multiple different scenes using the same settings with the VAF filter and without the VAF filter. For the first scene, I tested two lenses : a 35mm and a 10-18 Ultra-wide zoom meant for EF-S. Since I have the original VAF-5D2 filter and not the updated VAF-5D2b filter specifically designed for wide angles, I expect the performance of the VAF to be less using the 18mm focal length. Since I am using an EF-S lens on full frame, I expect the corners to be blurry and with vignetting, so I will use center crops only for comparisons. All of these shots are within 30 seconds of each other.

Here is my exposure for this test. Note the picture is at F/7, but I later changed this to F/8. The RAW zebras were only green in the sky, not green and blue like pictured.



Here are the tests. I added labels to aid in comprehension :

Click on each of these photos to see them in full size.










Initial conclusions :

Full frame video has now no aliasing when viewed at 100%. An interesting side-effect is the reduction of compression artifacts. This can be viewed clearly in the grass in both full frame tests. The reduction in sharpness is (to my eye) due to the elimination of false sharpness created by the aliasing. I most definitely think that the resulting image with the VAF filter is better when using FF.

Crop mode is straight up not usable. The sharpness reduction is dramatic. The shots look out of focus, but I can assure you I focused using the 10x liveview and autofocus on the central part of the image multiple times. I even re-did the whole test again because when I viewed the footage on the computer, I thought I made a mistake. We can conclude that using the VAF in conjunction with crop mode will yield poor results at best.

More testing :

After this initial test, I set out to test out different scenes to see the actual difference in real world tests. Since this filter was not intended for crop mode video and just for full frame, the later tests are all tested on full frame with the 35mm F/2 IS USM.








You can make your own observations. Here are the original files used in the test.


More conclusions :

With the further testing I did, I can say with confidence that for FF video in RAW on the 5D2, the VAF filter does its job very well. It reduces aliasing and compression artifacts to almost invisible levels when viewed at 100%. Crop mode is not to be used with the VAF filter. I do not think it was designed for that so I cannot fault it in that aspect.

Hope you found my tests useful !

ilia3101

Thanks so much. Always wanted to see what VAF-5D was actually like at the pixel level, in crop mode. Literally been waiting years for this test.

And it's just as I expected, vertical blur, to compliment the already exisitng horizontal binning.

ALso great to see a direct comparison in raw, without crop. Looks really good.


And what software did you process raw in? Looks like Adobe?

Volumetrik

Quote from: ilia3101 on April 23, 2020, 07:05:52 PM
Thanks so much. Always wanted to see what VAF-5D was actually like at the pixel level, in crop mode. Literally been waiting years for this test.

And it's just as I expected, vertical blur, to compliment the already exisitng horizontal binning.

ALso great to see a direct comparison in raw, without crop. Looks really good.


And what software did you process raw in? Looks like Adobe?

MLV mounted with MLVFS. DNG's then imported in AE and processed with ACR. Usually with reduced highlights and slightly boosted shadows.

The last tests have CineLog-C applied to them in ACR. I then apply an ARRI Log-C to Rec709 Lut using Lumetri.

Luther

Great test @Volumetrik. Can you just repost these images on a better host? This "ibb.co" thing only works with javascript enabled. I suggest using pictshare instead...

Volumetrik

Quote from: Luther on April 23, 2020, 08:48:55 PM
Great test @Volumetrik. Can you just repost these images on a better host? This "ibb.co" thing only works with javascript enabled. I suggest using pictshare instead...

Thank you ! I'll fix the embedding, I did not know it was a problem with javascript disabled !

reddeercity

@ Volumetrik I have a small request , can you try the VAF-5D with the crop_rec 4k 3x1(4096x590 = 4096x1770) preset and post results please ?
I'm interested to if can completely remove the line skipping artifact as good as the 3x3(HD) .
Since you use A.E. with ACR , just load the mlv in to mlv app & select 3x vertical stretch and export to fast pass cdng.
This will export @ the source bit depth & set a flag for 3x1 pixel for proper display (e.g. info says 4096x590 but displays 4096x1770)
and will not bloat the file size (like mlvfs), so 10bit read 30-40 % less then 14bit.

ilia3101

@reddeercity Agreed, might actually be the most interesting.

@Luther Has embedding improved for you? I changed it.

Volumetrik

Quote from: reddeercity on April 26, 2020, 01:49:36 AM
@ Volumetrik I have a small request , can you try the VAF-5D with the crop_rec 4k 3x1(4096x590 = 4096x1770) preset and post results please ?
I'm interested to if can completely remove the line skipping artifact as good as the 3x3(HD) .
Since you use A.E. with ACR , just load the mlv in to mlv app & select 3x vertical stretch and export to fast pass cdng.
This will export @ the source bit depth & set a flag for 3x1 pixel for proper display (e.g. info says 4096x590 but displays 4096x1770)
and will not bloat the file size (like mlvfs), so 10bit read 30-40 % less then 14bit.

Alright, i'll try that out soon ! I'll post the results here. It's this build right ?

Dec 31 2019
Experimental build: 4k(4096) 3x1 Anamorphic ,  2784x1160 1x1 @ 24.00fps & FHD 3x3 (1856x774) @ 48p


Quote from: ilia3101 on April 26, 2020, 02:23:07 AM
@reddeercity Agreed, might actually be the most interesting.

@Luther Has embedding improved for you? I changed it.

It certainty changed for me !

[img width=640]

I'll use the tag from now on. That's a much more elegant solution.

Volumetrik

I did some preliminary testing earlier today and I am getting a ridiculous amount of vertical lines.



I followed the procedure listed on the 3k/UHD thread, am I doing something wrong ?

They do seem to fade towards the end of the video. I recorded 5 seconds of each configuration and the last frames are better, but still have them.

Bender@arsch

Quote from: Volumetrik on April 27, 2020, 10:35:50 PM
I did some preliminary testing earlier today and I am getting a ridiculous amount of vertical lines.

You need to wait up to 30sec in crop mode before you start recording. It's still bleeding edge.

Volumetrik

Quote from: Bender@arsch on April 27, 2020, 11:26:09 PM
You need to wait up to 30sec in crop mode before you start recording. It's still bleeding edge.

Thanks for the heads up. I'll try waiting 30s before starting the recording when I re-test tomorrow !

Luther


yourboylloyd

Ahhhhh I've been waiting for this test!!

This is exactly what I needed to see. The VAF filter in the 5D2 is amazing for full frame! Those branches just melt away. I also now understand why Reddeercity and I disagreed so much on the use of the filter. That filter is unusable for his crop recording, and I never use crop mode.

It really gets rid of the aliasing... Man I'm jealous. I wonder what the difference between the VAF1 and VAFb is now.

Amazing test! This is proof that I needed to get a 5D3 for the filter and extra vertical resolution, but ever since the EOS R5 was announced, I'm just going to save the money and buy that no matter how much it costs.
Join the ML discord! https://discord.gg/H7h6rfq

Volumetrik

Here is the test requested by Reddeercity for the 4k(4096) 3x1 Anamorphic mode.

Setup is exactly the same as the earlier posts. I did four different scenes to get some variety for detail, depth of field, focus transitions and patterns.

The organisation of images will be with VAF on top and NO VAF at the bottom for each pair of images.















Here are the files as usual.


Quote from: yourboylloyd on April 28, 2020, 04:50:19 AM
Ahhhhh I've been waiting for this test!!

This is exactly what I needed to see. The VAF filter in the 5D2 is amazing for full frame! Those branches just melt away. I also now understand why Reddeercity and I disagreed so much on the use of the filter. That filter is unusable for his crop recording, and I never use crop mode.

It really gets rid of the aliasing... Man I'm jealous. I wonder what the difference between the VAF1 and VAFb is now.

Amazing test! This is proof that I needed to get a 5D3 for the filter and extra vertical resolution, but ever since the EOS R5 was announced, I'm just going to save the money and buy that no matter how much it costs.

Yes for full frame video I am extremely happy with the performance of the VAF filter ! I wanted the VAF-5D2b but these are so rare its ridiculous. I've been waiting a year for one to pop on eBay.. That R5 looks to be an absolute monster !

reddeercity

Quote from: Volumetrik on May 04, 2020, 10:28:20 PM
Here is the test requested by Reddeercity for the 4k(4096) 3x1 Anamorphic mode.
Thanks , Yes it looks like the VAF does help a lot , very surprising but it looks like at a cost of sharpness a little bit .
What lens did you use ?


Volumetrik

Quote from: reddeercity on May 05, 2020, 06:46:19 AM
Thanks , Yes it looks like the VAF does help a lot , very surprising but it looks like at a cost of sharpness a little bit .
What lens did you use ?

Yes that is my conclusion as well. Overall sharpness is slightly reduced, but jaggies and aliasing are pretty much all gone.

I used a 35mm F/2 IS USM between f/7 and f/9 with no filters.