F55: Center Scan vs Full scan test with same framing

The video makes for an excellent demonstration of effective use of center crop mode. Thanks for this Doug.

And again I will say, this round of feature additions through firmware is quite something.
 
harddrive said:
Yes, I also think the top one looks sharper than the bottom. I did wonder at first if the higher contrast was just giving an impression of sharpness, or there was more edge enhancement, but there does seem to be more genuine detail. The more you look, the more obvious it becomes.

I also agree that for the purposes of a test it's important to know what's in the plane of focus, otherwise you can be comparing two things, find them the same sharpness - but it's because one is soft due to not quite being in focus, the other due to lower resolution of the imager.

[EDIT - and after posting I saw the comments on page 2..... Glad to see I got it right! Promise I didn't cheat.]
Click to expand...

The test I didn't do but should have done was also shoot a 4k full scan version and crop out the center in post. It would have proved if the 2k center crop is as every bit of resolution we should expect from 2k effective pixels but because the 2k full scan is actually derived from a full 4k sensor it inherits a sharpness you just otherwise could not get from 2k effective pixels. Makes sense.

I also agree about the focal plane issue, its a bit hard to judge images when there is a narrow DOF but I also wanted to shoot something a little more real world then just a brick wall. So I shot some bush and balls. :huh:

In the two tests, the red ball is the point of focus in the Jingle Balls video, and the three little yellow/gold leaves are the point of focus in the Frozen Bush video.
 
Last edited:
The clincher to me was the silver ball with the red glitter bands. On the top image there is far more texture to the red bands than on the lower image. Having seen that, there are other slightly more subtle differences.

Is the centre cut 1920x1080 pixels? If so, it stands to reason that after deBayering it won't be as sharp as an 1080 image derived from the whole sensor.

I think the feature will be quite useful if RAW is highly important to a user - but 4K RAW is impractical for them. The lens aspects may be a curse or a blessing. A curse if you've got s35 glass and want to keep the w/a ability, a blessing if you've got easy access to s16 glass! :)
 
The center scan is 2048 x 1080 and so is the 2k full scan.

The two modes are really not too far off from each other which is the great news. I was really worried about some kind of aliasing issue like there is with the HFR modes. phew.

This new feature is great for those who want to double the focal length of their existing 35mm glass, or use super16 glass, or even want to use B4 2/3" lenses as with an 2x optical doubler built-in which I reckon will yield a 3000 x 1600 frame which you can easily derive 2k/HD from..
 
I have to agree with Dennis that the crop mode is a little bit softer if you look at freeze frames. Probably not softer enough to matter a whole lot in the final edit and delivery, but it is softer. Here are two more tests I did tonight.
Freeze frames were grabbed directly from the SxS cards with Content Browser.
You really need to download them and bring them into a program like Photoshop to see the difference.

http://www.vortexmedia.com/crop_test.html



The odd thing is that I don't see any difference in the actual video files in Premiere, -- as shown with my previous examples. They look identical there, and that's what counts the most to me.
 
Last edited:
I definitely see a subtle difference in the original MXF video files via Sony Content Browser, as well as the ProRes files I transcoded from the originals using FCP X.

After your recent post however I did go back into the Content Browser to export frames directly from there instead of FCP X and I agree that the shots look slightly less different than the ones I posted earlier. No idea why FCP X would amplify the difference more than Content Browser but then again it is Apple :)

Here are the tree frames directly from Content Browser:
content browser export full scan
content browser export center scan
 
Barry, I've already essentially done that. http://www.vortexmedia.com/crop_test.html
There's no question that the crop mode is a little softer when you're looking at frame grabs. Dennis and I have both demonstrated that independently, so unless you can show us a test that says otherwise, I'm going with what I've seen with my own eyes. But I wish you were right.

Fortunately, for reasons I can't explain, the softness of the crop mode that is apparent in frame grabs isn't evident during editing. When I put both files on a timeline, I can't detect any difference whatsoever. If I add a moving wipe between the clips, you cannot see the line move -- and that is always my standard when testing stuff like this. But that doesn't change that fact that there certainly is a slight difference when looking at frame grabs straight from the original video file.
 
Last edited:
Regarding softness and center vs full scan I think my test is a little more subjective because I did not shoot anything flat, and the actual focal plane could vary to a small degree making portions of the image look softer than others. I just wanted to shoot "real life" bush and balls if you know what I mean.

Having said that, the DOF does get narrower the closer you get to the actual subject and since the center scan shot is done from around double the distance of the full scan then it actually means my DOF was reduced drastically for my full scan shots. So in fact the full scan images should have more focus error than the center scan test shots.

In the end no one test really proves anything. As I learned through my testing and discussions with Doug it now seems for whatever reason my footage ingestion through FCP X actually seemed to have exaggerate the differences even more than if I had put the same original files through the Sony Content Browser. How the hell does one explain that? But that just proves that we can get different results at any point in the chain from shooting straight through to post.
 
Last edited:
Barry_Green said:
In theory, the crop mode can't be softer, in terms of actual resolved imagery.
Click to expand...
The way I understand it is that for full frame 2K, the sensor has 4096x2160 photosites at it's disposal. Even with the simplest deBayering, that means 2K x 1080 2x2 Bayer blocks. Since each block gives unique values for each of R,G,B directly, that means it will have no problem reaching full 2K resolution.

In centre crop, from what Dennis says, it's using a centre 2048x1080 window of photosites - a quarter of the total. To get the final image, it's necessary to do a full deBayer on that - which therefore means it CANNOT have full 2K resolution. With good algorithms, the normally accepted figure is 80% of the sensor dimensions in each direction - so I'd expect it to be around 1600x850 (around 850lpph).

That figure would predict pretty good resolution performance - but not quite up to what the 1080 system is capable of, the difference is certainly likely to be noticeable. And isn't that exactly what we are seeing? Pretty good performance - but unquestionably not up to "full" 2K?

Dennis, I suspect that if you do use a chart such as Barry suggests, it will be possible to quantify the difference. I would predict the full sensor 2K to easily make 1000lpph, the centre crop to be somewhere close to 850lpph.
 
Barry_Green said:
If I wasn't clear, let me clarify here -- I'm talking about HFR. Not about general 2K, not about full-sensor downrezzed to 2K, but specifically about it sampling 2K raw, where it uses 2x2 blocks and then skips the next 2x2. I'm interested in seeing how the sensor crop system works as compared to that, where I suspect the sensor crop must be noticeably, if not substantially, superior.
Click to expand...
OK, I don't dispute any of that.

I'm understanding Dennis's tests were comparing an XAVC 2K frame off the full sensor, and comparing that to centre crop and deBayered 2K, and I thought that's what you were also referring to.

I would agree that 2K RAW centre crop should be far better than 2K RAW derived via skipping off the full sensor.

[EDIT - Looking back over the thread, Dennis, am I right in what I was assuming above? I've got myself wondering now.....?]
 
Last edited:
Yes my test was 2k Full Scan (it means 2k derived from full 4k sensor with full super35mm field of view preserved) vs 2k Center Scan (refers to camera cropping out 2k window from center of full 4k sensor resulting in 2x crop factor).

I agree with harddrive that in theory, technically the center scan would have to be a bit short on resolution compared to the full scan mode.

On another forum it's been suggested that the softness comes from the fact I used f8-f11 on the shots instead of the sweet spot of the lens, but I don't agree with that much.
 
Thanks for the tests guys, I think we can safely conclude that at least now HFR can be shot on the cameras without problematic artifacts. The crop factor is frustrating, and the resolution obviously falls a fair way short of the 4K files, but given the 'dreamy' aesthetic HFR creates anyway (and that the detail is considerable regardless) I think things will be okay overall - though the Epic and upcoming Amira are clearly the stronger offerings for HFR (if we write off the 2K OPLF filter as too impractical for general use - though for anyone using the F5 without the raw recorder, I think it's a sensible investment).
 
My tests have NOT been in HFR mode, nor have I ever mentioned that mode in my tests or posts about the crop mode. I have no idea whether there would be a difference or not, but I suspect not.
I've done all the testing I intend to do because I've already come to my own conclusions about this shooting mode.
 
Grug said:
Thanks for the tests guys, I think we can safely conclude that at least now HFR can be shot on the cameras without problematic artifacts.
Click to expand...

No not yet because neither Doug or I have shot anything in the center crop HFR mode. Our tests are all at regular frame rates. We know in full scan HFR mode the camera's are pixel-binning to read at very high speeds, increase light sensitivity to deliver noise freeish image quality but at the small expense of some aliasing on edges. A 2k OLPF is available to address this but it quite softens up the overall image, so its a trade-off whether to use or not. It will depend on the individual scene.


Barry_Green said:
In slow frame rates, and assuming that the sensor is indeed using a 2x2->1 pixel system, then yes that's the expected results. But I would expect that the HFR center crop should blow away the HFR full-scan. Is that not available yet?
Click to expand...

Again not tested by Doug or I but it will really depend on how Sony has built it. Can they really just isolate reading the center of the sensor without having to read the entire 4k sensor and then crop? Who knows, but a center crop HFR test with some diagonal lines and bit of motion will easily reveal this. I'd be hard pressed to bet my money either way. Who knows what they have done.

Barry_Green said:
Well I absolutely agree with it as a *potential*. Diffraction is an absolutely resolution-robbing phenomenon, and shooting f8 or f11 on a 1/3" camera would destroy the resolution of the image entirely. On a 1/3" camera you can use up to f4 before diffraction sets in. On a 2/3" camera (which should be somewhere around the sensor patch being used in the center-crop system) you should probably be able to go to f/10 before diffraction becomes an issue, so it may or may not have affected your test. Simple way to know is to download that ISO 12233 chart and shoot it at f/2.8 and also at f/11 or f/16. If there's a substantial difference in sharpness, you'd at least know to avoid those smaller apertures.
Click to expand...

The lens I used was a Nikon 85mm f1.4 AI-s lens. It remains one of Nikons sharpest lenses ever produced and I had once shot some charts with it and did not notice any diffraction until f16. All lenses are different but this one is truly special. If you looked at a center crop shot with it at f4, f5.6 and f8 it would be very hard to see a huge difference. At f11 I admit there is a bit of reduction in sharpness but it is very small.

I really don't feel like repeating my test at f4 / f5.6 due to laziness at this point. But perhaps someone with a $27,000 lens wants to give it a go and see if the lens aperture really makes all that much of a difference. I think we all still agree that a 2x2 pixel > 1 pixel image is going to have a bit more sharpness over a 2k de-bayered to HD image.


Barry_Green said:
Okay, going back into this:

According to this schedule, December 2013 was supposed to bring two separate features: Center Scan Mode for S16mm Lenses, and HFR Center Scan Mode.

It is my understanding that these are two separate features. So my question is: in the tests that have been conducted, have they been done at 24fps? Or in HFR? And if they have not been done in HFR, can we please see an HFR comparison?
Click to expand...

Yes these are totally separate features. October firmware delivered the disappointing full frame HFR mode with pixel-binning and thus aliased images.

December firmware delivered center scan mode for s16mm lenses and also HFR center scan mode.

If I can snap out of my laziness from all the turkey, egg nog and sweets I've been eating this wonderful holiday season I will do my test again and include HFR and maybe even drop down 1-2 stops to satisfy the glass-philes.

Doug Jensen said:
My tests have NOT been in HFR mode, nor have I ever mentioned that mode in my tests or posts about the crop mode. I have no idea whether there would be a difference or not, but I suspect not.
I've done all the testing I intend to do because I've already come to my own conclusions about this shooting mode.
Click to expand...

Doug, since you have a $45,000 lens kit would you be willing to shoot a chart with only one lens in both center scan and full scan modes but only move the camera to get the same framing all while only using the sweet spot of the lens (about 2 stops over lower f-stop setting) so we can put "lens at question" issue to rest?
 
Last edited:
Thanks, but no thanks. I'm going to take a pass on doing any more testing on this. As you and I both know, it is a pain in the ass to try to do it right and eliminate all the variables and potential problems. We've done are part. Let some one else take the baton and run with it if they feel it is so important. I've got the answers I was looking for and have now lost interest in going further. Besides, I already pretty much did the test you want already anyway: http://www.vortexmedia.com/crop_test.html Both tests were done with the same lens and the same f-stop, just moving the camera to try to keep the framing as close to identical as I could.

But I'm in agreement with you that the lens is probably not a big factor anyway. And what are people going to do, run out and buy new lenses to use center scan? If a lens offers certain performance in full frame, it out ought to be the same in center scan. Why would it be different? If you like your lens for full scan, you'll like it for center scan. What more do you need to know?
 
starcentral said:
...... but it will really depend on how Sony has built it. Can they really just isolate reading the center of the sensor without having to read the entire 4k sensor and then crop? Who knows, but a center crop HFR test with some diagonal lines and bit of motion will easily reveal this. I'd be hard pressed to bet my money either way. Who knows what they have done.
Click to expand...
The way sensors work, the options available to Sony (or any manufacturer) for a given sensor are relatively limited.

Basically, for any single frame, a single line can either be read - or not. Given that, any of the photosites then along it can be ignored if desired. In the case of centre crop, then half the lines never get read at all (the top and bottom quarters), then half the photosites on each row get ignored - the first and last quarters. That is very straightforward.

If full sensor 2K RAW does as Barry says (and I've every reason to believe that it is so), then the read requirements on the chip are exactly the same - though very different in nature. It's still only reading half the lines - but it's now read two, skip two instead of skip 540, read 1080, skip 540! Of the ones read, it's keep two, skip two etc - rather than just keep the central half.

The obvious big difference between the two will be in angle of view for a given lens. That said, it's impossible to believe that the gaps in the read pattern in full s35 frame mode can't but make a negative impact on performance compared to reading 2048x1920 photosites in a continuous block. Maybe not too much in resolution terms - but certainly in terms of aliasing. This is why theory strongly supports any view that centre crop mode can only be superior to a "full sensor 2K RAW" mode with pixel skipping. (And both inferior to reading all 4096x2160 photosites and deriving 2K via the 2x2>1 method.)

And that's where "real world" tests fall down. To make sensible conclusions you HAVE to use a chart, and it MUST be of a type that easily and clearly distinguishes aliasing from real detail - which means something like zone plates, charts with only horizontal and vertical lines won't do. Do the test with only "real world" images and it becomes impossible to sort out what is causing what - too many variables. (Though such are valuable for "how much does it actually matter in real life" - I'm not disputing that.)

And with a zone plate, the real beauty is that it sorts out lens issues. If you see true detail with aliasing starting at a certain resolution, the lens *MUST* be resolving past the limiting resolution of the system. If the lens was the limiting factor, then you'd never see aliasing in the first place.
 
Last edited:
Must agree with harddrive. People who think they are somehow learning something more practical about a lens by avoiding charts are always adding a wealth of variables that make real comparisons very questionable. By all means do "real world" shots to isolate specific issues ( highlights, backlight, DOF, etc,.) but charts are the best way to analyze performance and should be the starting point. Years ago when people first started using those 35mm adapters Phil Bloom wrote some glowing reviews of the original Letus adapter and praising its sharpness. I put it on a chart and showed it was terrible - not sharp in the corners till 5.6. Letus immediately reacted and improved their optics. but it was a classic example that charts have an important role. That's not to criticize Phil who never claimed to be scientific and whose tests also revealed a great deal of other info.

Wish i could afford an F55 to do the tests you guys are doing.
 
Someone at Sony has told me that the F's do not using line skipping to perform full frame HFR or center crop at regular frame rates, and I suspect they are not using it for 2k RAW either. Line skipping would indeed create some aliasing (like we see on DSLR's) but it will not improve light sensitivity while reducing noise required to record at very high frame rates like pixel-binning would (which is what Sony told me is the method that the cameras use).

Regarding the 'HFR center crop' neither Doug or I have tested it yet, but surely the cat is probably already out of the bag over on the SCF.

As for real word tests vs test charts I think both are useful to look at but feel the final verdict lies in the results of "real world" images not bench tests. In the end I'm sure more tests will emerge shortly.
 
starcentral said:
No not yet because neither Doug or I have shot anything in the center crop HFR mode. Our tests are all at regular frame rates. We know in full scan HFR mode the camera's are pixel-binning to read at very high speeds, increase light sensitivity to deliver noise freeish image quality but at the small expense of some aliasing on edges. A 2k OLPF is available to address this but it quite softens up the overall image, so its a trade-off whether to use or not. It will depend on the individual scene.
Click to expand...

Doug Jensen said:
My tests have NOT been in HFR mode, nor have I ever mentioned that mode in my tests or posts about the crop mode. I have no idea whether there would be a difference or not, but I suspect not.
I've done all the testing I intend to do because I've already come to my own conclusions about this shooting mode.
Click to expand...

Oh, I was under the impression that Sony had already stated that the crop-mode HFR wouldn't require any line-skipping or pixel-binning - I thought that was the whole reason for most of the excitement for it.

If cropped HFR mode does use the full sensor readout, then your examples should be exactly the same as HFR - if not, well then we're back where we started as far as HFR is concerned.
 
You must log in or register to reply here.
Back
Top