Finally a 7D resolution test...

[Duke]

A guy posted a 7D mov file directly from the camera, with sharpness 0, color 0, etc. of a proper resolution chart. The only thing he didn't specify was what dpi he printed it at. (I printed one at 1200dpi while awaiting my 7d.) He used the kit 18-135 L lens. Also, the Iso was 1600 which had to add a little grain with the gain.

It here:
Canon 7D - Resolution test on Vimeo

If you Log in you can download the clip from the lower right corner.

The poster said he thought the resolution was 1200v x 900h. When I looked at one frame my impression was ~800v x ~700h. Not the 480p they've been claiming on Scarletuser. Of course they may be a little biased. :hv20-smilie15:

First, a sensor 1920 pixels wide doesn't mean you have 1920 resolution. Usually we are talking about line pairs. If understand it correctly 960 would be the maximum if it was perfect and nothing reaches perfection.

Then you'd have loss due to debayer, and he lost a little more due other factors, like gain, hand held (the image shifts), etc.

Another factor I thought was important was that the angled black boxes do NOT show aliasing artifacts! Moire patterns are shown at certain frequencies.

I've posted three captures where I've blown up sections 300% to show imperfections. Due to the gain and hand held I think under ideal circumstances it will be a little better than this. I don't 'think' Vimeo has recompressed the file, but I don't know for certain.

[Duke]

Are there any other resolution tests out there yet?

[CorbenT]

Someone should do a test with a super sharp canon/nikon/pentax lens and upload that. I have a feeling it would be sharper in the corners than the standard zoom which the 7D comes with. Its an interesting test, but I'm sure that it won't change peoples' views towards the camera, as most people want to achieve the elusive film look, which high sharpness is not often a part of.

[Halsu]

Someone should do a test with a super sharp canon/nikon/pentax lens and upload that. I have a feeling it would be sharper in the corners than the standard zoom which the 7D comes with.

A super sharp lens would make the already pretty ugly color moiré look much worse...

As far as resolution goes, the chart looks decent up to roughly 700-800 lines (350-400 line pairs), after that all detail that's there is moiré ridden and should have been filtered out if the camera was designed properly.

So, it's better than 480p, worse than 1080p.

[1]

...and this was shot hand held? That's a new one for a rez test!

[Eugenia Loli-Queru]

The guy needs to use a better lens, use good light, low ISO, and a tripod. Then, we talk again.

[Koppel]

Hand held and ISO1600 - that's not right.

A good lens can narrow down the variables but I think we can get decent information with the kit lens as well.

[rhervag]

aaah 1/30 ruins the beauty of handheld precision imho :D
that's a funny one !
obviously high ISO is to be avoided. for tests, fast L lenses please.
it's about time some people see how much the glass makes a tremendous difference in picture quality. sensor, resolution etc.. all are pointless specs when shooting through a bottle of coke.
many have been raving about pbloom's and others footage forgetting their fine ZEISS and primes costing far more than the camera... prepare for serious disappointment when looking for such results with the kit lens. it is slow and cheap.

[1]

Hear hear!

Well stated, rhervag!

[I C]

aaah 1/30 ruins the beauty of handheld precision imho :D
that's a funny one !
obviously high ISO is to be avoided. for tests, fast L lenses please.
it's about time some people see how much the glass makes a tremendous difference in picture quality. sensor, resolution etc.. all are pointless specs when shooting through a bottle of coke.
many have been raving about pbloom's and others footage forgetting their fine ZEISS and primes costing far more than the camera... prepare for serious disappointment when looking for such results with the kit lens. it is slow and cheap.

I find this statement somewhat ironic when in your signature you say that you use Canon FD primes on a 5d mark ii. The only way to use FD primes without losing infinity focus is to use an adapter, which degrades the image quality even if you have the expensive Canon-brand ones.

You are right though, the lens plays a HUGE role. I constantly see people crank out better photos with a Rebel and nice L glass than a 5d or other "professional" camera with crappy stock/zoom lenses.

[1]

I find this statement somewhat ironic...

Maybe, but he ain't posting rez tests! :hv20-smilie77:

5D MkII with a Phoenix, $160 lens here......also no rez tests from me though! :hv20-smilie81:

[I C]

Maybe, but he ain't posting rez tests! :hv20-smilie77

lol. True, true

I am curious though, are you using a glass adapter with those FD lenses, or one without glass and just forgoing infinity focus? I toyed with the idea of an adapter to use all the FD primes I had bought in anticipation of the GH1.

[Erik Bien]

...and this was shot hand held? That's a new one for a rez test!

Personally, while I think a lock-off shot would be "correct" if we were talking about the camera's still capabilities, a bit of camera movement makes it much more relevant if we're evaluating the camera's video capabilities ...

[Duke]

I think locked down on a tripod, a better lens, a properly printed chart, a lower ISO and a monitor for focus would all help the rez test accuracy.

We need someone that will do the test properly. Any one have a 7D handy?

[rhervag]

I C you are right of my being ironic ! //OFF thread _
yes a simple FD>EOS ring without lens is fine for macro work but loses focus after a short distance so no infinity right.
FD lenses can also have the rear removed and modified to an EOS mount : a costly difficult operation better carried out by a profesional with skill and toolset. worth only for expensive L glass and specialties like tilt-shifts. Vincent Laforet has the amazing FD 7.5mm listed as such for 'reverie'.
then there's at least 4 differents adapters with optical correction (Canon's own super expensive has a protruding end : doesn't work with primes as it was designed for longer zooms ; the Kenko one supposed to have good glass is a rarity nowadays ; and then for the most of us we find the cheaper ebay ones from BOWER/HOYA with generic lens as well as the newly surfaced AF-confirm from jinfinance : both i've used with mixed results ; they act as a 1.25x teleconverter and can NOT handle wide apertures so forget about f1.2-f1.4 etc... stop down to at least f2.8 for contrast and sharpness or assume some glow in highlights.
it seems everyone has been rehashing what a few photogs said after pixel peeping stills and some myth carried out onto FDs for videomaking.
*f2.8 is plenty shallow and more than enough before it becomes a nightmare to follow focus, at least on fullframe but did i mention the moiré and aliasing issues most complain about with their EF lenses are greatly reduced using such a bad glass setup... AH!=)
so on the other hand, if you like softness diffused images rather than razor sharp ENG style, you might be pleased to have such character in your bag of tricks...great for music videos with that virgin suicides look done instantly rather than with tons of postwork.
i know many love to spend hours 'magicbulleting' some neutral footage while i prefer to slap appropriate glass and filters to get it right optically onlocation to maximise codec usage.
Back to thread //

[thequads]

the chart aint bad at all. It's no scarlet killer, but def a killer of most camcorder/adapter combos.. with the exclusion of the ex1/adapter combo.

[antman]

Why do you think this is a camcorder and dof adapter killer? low light, and res?

[1]

Personally, while I think a lock-off shot would be "correct" if we were talking about the camera's still capabilities, a bit of camera movement makes it much more relevant if we're evaluating the camera's video capabilities ...

....yeah, as long as all other rez tests do the same. :hv20-smilie77:

It needs to be next to a locked down test, in order to have value; as it is, it's pretty useless....

Here; I did a quick resolution test with my 5D MkII & a Canon 70-200mm IS 2.8L lens (IS on), out of a moving car going 40km/h, whilst holding the cam in one hand, a small plastic Statue Of Liberty in the other, and at the same time brushing my teeth:

[Duke]

Barry Green just posted a trumpet rez test on DVXuser.
http://www.dvxuser.com/V6/showthread.php?t=186334

He compares it to the HMC40, side by side. The funny thing is he doesn't label which side is which. I eventually figured them out, but the really funny thing is the odd star moire pattern that he complains about is on the left side of one, and the top and bottom of the other.

Here's the chart on the 1080 test. Theres some weird stuff from both cameras, but I read the chart a little higher than he did. On many trumpets I think it reaches 800-900 lines, just a little better than the other test, which makes sense. Also thats about the same level as the HV20 test when using tape when I was testing HDMI.

http://dvxuser.com/barry/7D-HMC40-1080-Mid-Trumpets.jpg

[thequads]

Why do you think this is a camcorder and dof adapter killer? low light, and res?


yeah both of those.. it doesn't quite kill all the camcorders in terms of resolving power, for example, the canon hfs series resolves more, but when you add the amazing lowlight and dynamic range.. it's hard to beat.

[rhervag]

Here; I did a quick resolution test with my 5D MkII & a Canon 70-200mm IS 2.8L lens (IS on), out of a moving car going 40km/h, whilst holding the cam in one hand, a small plastic Statue Of Liberty in the other, and at the same time brushing my teeth:

:))) not bad indeed kudos for the clean teeth ! how about IS off ? :hv20-smilie45:

[1]

how about IS off ?

Long live Canon's IS: :)

[1]

...it's hard to beat.

I find a dedicated Video cam easier to use though...shape/form/button layout and all.

Vid cams won't go anywhere just because still cams do video.

[thequads]

I find a dedicated Video cam easier to use though...shape/form/button layout and all.

Vid cams won't go anywhere just because still cams do video.


yeah, I guess the vdslr isn't good for many situations. I for one am sticking to the video camera adapter combo for now.

[Erik Bien]

Here's the chart on the 1080 test.

Oof! :eek:

Not only is there mucho aliasing happening on the left there, look how different it is from one trumpet to the next — that's gonna be murder on moving fine detail.

[1]

...that's gonna be murder on moving fine detail.

So what's the deal?

Are these cams deliberately crippled with poor video processing? I mean, the HMC40 looks pretty darn amazing; and that's from a 1/4" chip?????!!!!!


Are these chart tests NOT really giving us a proper interpretation of what actual footage looks like?

[Duke]

I noticed Barry didn't post the ISO, the lens, the f-stop or anything necessary to judge the image. Some people said the moire is better with sharpening off.

[Erik Bien]

Are these cams deliberately crippled with poor video processing?

In my inexpert opinion, I'd say yeah: whether they're skipping rows of the sensor, as some have said, or 'pixel binning' is I think what Canon claim, there clearly isn't enough low-pass filtering happening on the video signal (makes sense when you think it needs to also take sharp stills with much higher pixel counts). :(

[Ian-T]

So let me ask....would a different lens make a big difference in this recent test?

[rhervag]

for what its worth, i suggest that the lens is important indeed.
what is desireable for stills is not always a good thing for shooting video. the sharpest glass will optically resolve too much detail that produce more aliasing artefacts once lines are skipped than a cheaper lens that optically blends details offering a somewhat smoother pattern on the sensor with better averaged end results after line skipping since there is no proper filtering.
this gets especially obvious on very detailed locked shots where any little shaking alters only slightly the position of pixels from the picture projected.
shooting a grid in windy conditions from a sturdy tripod can create horrid busy backgrounds due to micro-shakes alterning the shades on the fixed pattern.
the only way to get rid of that awful aliasing effects is to defocus slightly so that no smaller detail can fall inbetween skipped lines.

[1]

...the sharpest glass will optically resolve too much detail that produce more aliasing artefacts once lines are skipped than a cheaper lens...

Ha! Interesting concept. Would be kind of interesting to find out -through examples- if this holds true, and that indeed some old-ebay-glass or just cheaper new stuff yields better, overall pleasing, video imagery!

[rhervag]

@1 that's why i like FDs but please don't tell anyone... keep'em cheap orphans and obsolete on ebay until i got'em all :)
they were plagued by mega pixels photogs to be unusable due to shitty optics in the adapters causing softness and IQ degradation... which has been a blessing so far to add a vintage feel to my filming compared to the somewhat sharp and sterile kit lens.

[Duke]

By that token you could use modern glass and just defocus a bit.

[rhervag]

By that token you could use modern glass and just defocus a bit.
yes indeed. but it's an all different look and feel to the image, even between tamron sigma and canon current offerings. obviously i do use latest canon L for assigments but for personnal stuff i indulge in manual land chasing older classics.

[Duke]

There's practically a cult following of people gathering up old 16mm film glass to use with the 2/3" Scarlet.

They claim it's higher resolution than non-film glass, and then claim it's less sharp/diffused, which I always find amusing.

I wish my 7D had arrived. I have a Canon 70-200mm f2.8L lens sitting here right now that I'm going to drop off at for a friend. I could test it in an instant to see if the lens was the issue.

[Duke]

This is probably the best explanation I've ever seen of lenses in general and Canon lenses in particular. :hv20-smilie31:

http://photonotes.org/articles/beginner-faq/lenses.html

[1]

This is probably the best explanation...

THAT'S in-depth!!!!

But why exactly did you post this?

Here's the VERY cool CANON site that explains how they make the lenses:
http://www.canon.com/camera-museum/tech/l_plant/index.html

[Duke]

It was so in depth it could/should be its own sticky.

I put it here because so many people were asking whether the lens would make a difference in resolution, and what lenses would work with the 7D, that it started me thinking about such things.

I was also wondering about my c-mount lenses and whether they'd work on the 7D with an EOS adapter. (I have one pro c-mount lens that the MSRP was originally $3,000 on the Fujion site.)

From what was said on those pages it would, but would likely lose infinity focus without a teleconverter, which if used would lose a f-stop or so. Since it's an f1.8 lens that might not be too bad.

Feel free to move it.

[Duke]

Looking at it further, because the back focus is shorter on a C-mount, the c-mount lens wouldn't work well on the EOS camera even with a teleconverter and an adapter, but...

The good EOS glass will work well on a c-mount camera, like the Scarlet, but...
it will act like a 3.5 crop factor. A 10-20mm would be great.

[Duke]

I tell you, when I looked at those rez charts my faith in the 7D was shaken. I looked up the rez numbers of the A1 and HV10. I'm not happy that its less than the resolution of a 1440 1.3 PAR camera in its worst mode (24f), but its not so far off that they couldn't be intercut.

However, when I look at real life situations being shot with the 7D, I'm blown away. Barry Green made a comment that the 7D might best be used for interviews, but not at a wedding because the veil would create a moire nightmare. Then I saw this:

phyllis + adam // same day edit on Vimeo

Shot with two 7Ds and a 5D. They cut together well and I'm totally blown away. No moire, beautiful images, nice motion without wobble. The biggest artifact I saw was rolling shutter when the flash bulbs went off.

XH-A1 60i: (01/01/2007)
824.3 Lines Horizontal (MTF50); 810 Lines Horizontal = Limit@Nyquist
674.8 Lines Vertical (MTF50)

XH-A1 30F: (01/01/2007)
823.6 Lines Horizontal (MTF50); 810 Lines Horizontal = Limit@Nyquist
592.2 Lines Vertical (MTF50)

Previous Reported Result for XH-A1 24F:
810 Lines Horizontal (Limit@Nyquist)
590.4 Lines Vertical (MTF50)

Previous Reported Result for Canon HV10:
784.5 Lines Horizontal (MTF50)
704.8 Lines Vertical (MTF50)

Faith restored, I'm not going to cancel my 7D order.

[Ian-T]

I saw that last night. I loved that opening scene...it played out like a cinematic film.

One thing I’ve learned about resolution or perceived sharpness is that contrast plays a very important role. The test charts can say one thing but in real life applications both the 7D and the 5Dll looks much better because of that and other factors. I think that has a lot to do with having a much bigger sensor than all those other cams to begin with. Think about it, when you take a 1920x1080 video and shrink it down to a much smaller size it appears to be very sharp, detailed and the colors kind of POP even more. Well, that’s how I look at ALL of these DSLRs. They all start off at such a huge image before being dumbed down to what we see in the end codec. But because they started off so big and where condensed “down” to 1920x1080 they all seem to have that sort of POP that helps give them what I call a much more DENSE image than our standard consumer/prosumer camcorders. The contrast and colors look much better which helps with their overall perceived sharpness.

I guess you can look at it the same as oversampling. Like taking an HD image, scaling it down to SD....and getting a much better SD image than any SD cam can produce internally.

[Halsu]

phyllis + adam // same day edit on Vimeo (http://www.vimeo.com/6650684)

Shot with two 7Ds and a 5D. They cut together well and I'm totally blown away. No moire, beautiful images, nice motion without wobble. The biggest artifact I saw was rolling shutter when the flash bulbs went off.

That was a pretty darn well done wedding video... but that also means the photographers (and/or editor) skillfully avoided the pitfalls where such artifacts would show too much - the video would probably have looked OK artifact wise even if it was shot with D90.

[Halsu]

I guess you can look at it the same as oversampling. Like taking an HD image, scaling it down to SD....and getting a much better SD image than any SD cam can produce internally.

That would be the case if the cameras used the full sensor. But because of speed problems, they skip pixels.

This is the same as scaling HD to SD, but instead of averaging (like i.e. photoshop's bicubic scaling), you simply throw away all extra pixels (like i.e. photoshop's nearest neighbor scaling).

The result looks very crispy, but small detailed patterns will have aliasing, especially noticeable with moving images.

[Ian-T]

That would be the case if the cameras used the full sensor. But because of speed problems, they skip pixels.

This is the same as scaling HD to SD, but instead of averaging (like i.e. photoshop's bicubic scaling), you simply throw away all extra pixels (like i.e. photoshop's nearest neighbor scaling).

The result looks very crispy, but small detailed patterns will have aliasing, especially noticeable with moving images.The edge modern camcorders have over DSLRs (video mode) is that they can see or resolve what they see better. But if the end result looks better on the DSLR (while viewing on the PC) then I believe it will still look better on the big screen (especially because the picture comes from a 1920x1080 image to begin with).The thing is if one were careful with not getting any aliasing/moire etc in their shots while filming on the 7D then the resulting videos (as we’ve seen in many videos) can look much better than any of these consumer/prosumer cameras that can actually resolve their image better. I mean…lines of resolution and how you got the image is one thing. But once recorded to a similar size (eg. 1920x1080) all bets are off. It’s basically the same as your thought on dynamic range in post production.

[Halsu]

But if the end result looks better on the DSLR (while viewing on the PC) then I believe it will still look better on the big screen (especially because the picture comes from a 1920x1080 image to begin with).

As you say, it's the end result what matters - if we skip ergonomics and consider just the imaging features, there are pluses and minuses for both (depending on model):


DSLRs:

+ Shallow DOF for those who want it
+ Better low light performance
+ A big variety of lenses
- Aliasing artifacts
- Color moiré
- Nonstandard frame rates
- Worse skew
- Shallow DOF makes focusing harder
- No humanly priced long range zooms available

HDV:

+ No aliasing artifacts
+ No color moiré
+ Standard frame rates
+ Deeper DOF makes focusing easier
+ Built in long range zooms
+ HDV seems to be a better codec for post than what current DSLR's offer
- No way to get a really shallow DOF (without adapters)
- Poorer low light performance

DSLR's are good for low light work, and make for pretty portraits with blurry backgrounds. Their problem is wide angle shots, with sharp small details: venetian blinds, patterned fabrics, tiled roofs, brick walls... you have to choose your subject carefully, or you end up with very nasty problems.

HDV cameras are better for all-around work, but they suffer in low light conditions, and getting that artsy shallow DOF look is harder. But you can shoot pretty much any subject with them, without worrying about crawling nasties around your screen.

As far as color and dynamic range go, it depends on camera settings: even though a lot of the DSLR footage looks overly contrasty with clipping highlights, i assume this could be remedied by adjusting the camera's color settings (at least on my D90 this is possible). Detail wise, neither any of the HDV/AVCHD camcorders OR any of the DSLR's resolve near to what's possible to store in an 1920*1080 image, so the 1440 vs 1920 debate is pretty moot... also, most cameras use way too much sharpening by default, which makes comparisons even harder.

[lordtangent]

At some point (we are not there yet) HDSLRs will get proper low pass filtering and debayering (like the Red cameras). The paradigm shift the HDSLRs have introduced is a HUGE deal and things will only get better as the consumer cameras improve. If Canon were smart, there would get right on to creating a larger sensor "video camera" ASAP.

A larger, single sensor has a lot of advantages over any other approach to acquiring a moving digital image. (reasonably priced, long zoom lenses being about the only major weakness) Once they sort out the low-pass filtering and debayering and get it to "Red quality" level things will be pretty amazing.

In a lot of ways the 7D is clearly "good enough" for many purposes so it will do for now.

We live in very interesting times!

[Ian-T]

- Nonstandard frame rates
I think this only applies to the 5Dll though. Also the moire (I believe) is not a problem in 24p but 60p is the issue (and honestly not all the time....but when it does happen it can get ugly),

[Halsu]

I think this only applies to the 5Dll though. Also the moire (I believe) is not a problem in 24p but 60p is the issue (and honestly not all the time....but when it does happen it can get ugly),

Half of the world lives in PAL countries - we need 25p or 50p, which are pretty rare commodities with DSLR:s, at least for now. 24p and 60p are pretty much as useless to us as 30p.

Moiré is a pretty big issue in my opinion - not all projects comprise of shallow DOF closeups only... it's not a show stopper, but still pretty restrictive.

But as said, we live in interesting times, in the next year or two there's going to be a lot of cool new camera gear around to geek over here in Europe too ;-)

[Ian-T]

You do know that every 7D made shoots 24p, 25p, 30p, 50 and 60p? there is no PAL or NTSC cam...they all have these frame rates.

[thequads]

at first I was turned off by the moire and aliasing, but then realized that many cams already exhibit this and it's never been a big deal before. Unless the 7d has unusually bad moire and aliasing, i don't think it's much to sweat about.

[Halsu]

at first I was turned off by the moire and aliasing, but then realized that many cams already exhibit this and it's never been a big deal before. Unless the 7d has unusually bad moire and aliasing, i don't think it's much to sweat about.

It's never been a big deal before, because cameras that are designed for video or stills only do not have significant moiré problems - it's just the current, pixel skipping video capable DSLR's that suffer from this badly. It's a new phenomenon from end user perspective, but well known to camera manufacturers.

The reason is simple: practically all digital cameras (both still and motion) have a built in low pass filter that eliminates small detail before it reaches sensor, blurring the image slightly to reduce moiré and aliasing caused by "smaller than one pixel" detail. In DSLR's this filter is set up with the full still photo resolution in mind, and it works fine for that, distributing small detail over multiple pixels - but when just a part of the photosites of the sensor are used in video mode, the filter is way too weak, and doesn't work anymore. Interesting rainbow like artifacts will show.

In worst case example, in a moving image, a detail is visible in one frame, shows only in the red channel in the next frame, and falls between pixels and thus disappears in the next frame.... the resulting video can be, um, interesting ;-)

I did a quick search and found an example from 5DMKII:
Moire Test on Vimeo

Here's 7D:
720p with moire (Canon 7D pre-production camera) on Vimeo

[Duke]

HDV:

+ No aliasing artifacts
+ No color moiré

But you can shoot pretty much any subject with them, without worrying about crawling nasties around your screen.


I would agree with most of what you said, but not the above. HDV, and any digital camera, will suffer aliasing and/or moire with certain shapes and patterns at certain frequencies.

Even though the HMC40 looked nice and crip resolution-wise on Barry's chart, it still suffered both, but in different locations on the chart. Also, when I posted those comparison shots of different cameras a couple months ago they all showed alisasing on the curves, it was only a matter of degree. Its a matter of trying to show a curve by using a grid. You will almost always have some stair stepping.

Down rez-ing always produces more aliasing. Hopefully the Scarlet will have that figured out because it's files will almost always be down rez'd.

I do think it would be fair to say that the 7D sufferes more from aliasing than most, and as you say from the skipped pixel/binning issues.

Careful planning of your shots can avoid that, but the point of the wedding video is they shot the veils and patterned dresses without moire or artifacts being noticed (other than rolling shutter from the flash.) I believe the real life motion helped tone down the artifact.

[Halsu]

I would agree with most of what you said, but not the above. HDV, and any digital camera, will suffer aliasing and/or moire with certain shapes and patterns at certain frequencies.

Sure, artifacts can happen, especially when sharpened too much. But it's not even in the same planet as the problems with DSLR's severity wise. It also IS possible to make the OLPF strong enough to cancel all visible aliasing and moiré, but the image will be pretty soft - I.e. HV20 in cinemode is a pretty great compromise...


Down rez-ing always produces more aliasing. Hopefully the Scarlet will have that figured out because it's files will almost always be down rez'd.

No, no, no. Shooting at a higher resolution and then downrezzing is the prime way to AVOID aliasing while maintaining the highest possible resolution at final delivery size. It just has to be done properly.


I do think it would be fair to say that the 7D sufferes more from aliasing than most, and as you say from the skipped pixel/binning issues.

Careful planning of your shots can avoid that, but the point of the wedding video is they shot the veils and patterned dresses without moire or artifacts being noticed (other than rolling shutter from the flash.) I believe the real life motion helped tone down the artifact.

Yep - in most real life situations it's a non issue, just like rolling shutter is usually a non-issue.

But one should be aware of the potential problems because in worst cases they can ruin a shoot to the degree that the footage will be unusable.

[Duke]

No, no, no. Shooting at a higher resolution and then downrezzing is the prime way to AVOID aliasing while maintaining the highest possible resolution at final delivery size. It just has to be done properly.


It depends on the way its down rez'd. The way I heard it discussed with the Scarlet was using data from adjacent pixels. That sounds a lot like the binning described by Canon. (Combining data from several pixels.)

It didn't work too well on the 7D. If skipped pixels was used on the 5D, that didn't work too well either. Who knows if those rumors are accurate, but test images seemed to support those conclusions on the 7D and 5D.

In most cases a down rez creates a bigger jump in data irregularities by creating an unevenness of the data distribution. (This is the same thing that causes banding when color grading in 8 bits.)

Good 3D programs deal with this aliasing by edge detection and then antialias a limited area. That's too slow for a camera.

I'm not saying the Scarlet won't have it whipped. I said I hope they've figured it out because it is a problem. There are a few ways that work, and a bunch that don't.

[Ian-T]

Here's a cool workaround for the moire and aliasing issue. Defocusing slightly while filming and adding a slight sharpness in post works wonders. Sure, it's an extra step of work but it does work. Check out the rooftop of the building .

http://www.dvxuser.com/V6/showthread.php?t=186726

I'm sure some resolution was lost ...somewhere .....but please point that out to me because you can not tell just by looking at it.

[Erik Bien]

It depends on the way its down rez'd. The way I heard it discussed with the Scarlet was using data from adjacent pixels.

This is only a question with regard to the still-unknown "1080 RGB recording mode" Scarlet will be able to perform in-camera; debayering and scaling of the REDcode RAW .R3D files will be done RED One-style — in post, where you have several options to choose from (nearest neighbor, bicubic, Lanczos, etc.). The ability to separately choose and fine-tune your debayer and scaling options when converting from RAW is IMHO one of RED's biggest advantages.

[Erik Bien]

Defocusing slightly while filming and adding a slight sharpness in post works wonders.

It's effectively the same as using an older, less-sharp lens to blur detail beyond the aliasing threshold (seeing as how I have a soft spot for nice older glass and considering that lots of old manual lenses are available cheap, that's probably the route I would try to go).

[pascalbrown]

so we are upgrading our cameras and then downgrading the glass? Personally I find that a little amusing...

[Duke]

This is only a question with regard to the still-unknown "1080 RGB recording mode" Scarlet will be able to perform in-camera; debayering and scaling of the REDcode RAW .R3D files will be done RED One-style — in post, where you have several options to choose from (nearest neighbor, bicubic, Lanczos, etc.). The ability to separately choose and fine-tune your debayer and scaling options when converting from RAW is IMHO one of RED's biggest advantages.

No doubt. As you say, Red provides options in post. However, the issue is whether the Scarlet will ever suffer from moire or aliasing, and contrary to what Halsu asserted, it will (as do all digitally recorded images), and more so due to down rez-ing.

You'll have options in post, but to a degree everyone will have some options in post for every camera. What you can say is that Red gives you a much better chance of correcting it.

[um3k]

Going by these test results, I'll readily believe Canon's claim that it is binning pixels, but only in the horizontal axis. Vertically, there is no doubt whatsoever in my mind that it is skipping pixels.

[Halsu]

No doubt. As you say, Red provides options in post. However, the issue is whether the Scarlet will ever suffer from moire or aliasing, and contrary to what Halsu asserted, it will (as do all digitally recorded images), and more so due to down rez-ing.

No. All digital images do NOT suffer from (visible) moiré or aliasing. Only the ones that have too high frequency details or too sharp edges do.

Starting with a high resolution original taken with a camera that has a proper OLPF, and scaling it down in post with a proper algorithm is the best possible route to get a smaller sized image without visible aliasing or moiré, while at the same time maintaining the highest possible sharpness.

Downrezzing will not produce moiré unless done badly. The proper procedure may i.e. require blurring the image slightly before scaling etc. Just scaling the image down with your editing software is often not good enough...


You'll have options in post, but to a degree everyone will have some options in post for every camera. What you can say is that Red gives you a much better chance of correcting it.

No again - the moiré problems with current video DSLR's can NOT be remedied in post.

[Duke]

No. All digital images do NOT suffer from (visible) moiré or aliasing. Only the ones that have too high frequency details or too sharp edges do.

All digital cameras create aliasing, because it is a square grid trying to render curved shapes. As for the Red not suffering from moire when down rez'd, well, people using Red cameras are talking about it.

http://forums.creativecow.net/thread/242/2516



No again - the moiré problems with current video DSLR's can NOT be remedied in post.

The old fashioned way was to uprez an odd amount like 33%, then blur, then down rez. The blur creates some anti aliasing. Then sharpen a bit.

Here's an example of the same idea. Nothing is perfect, but it can save a shot.

http://www.dvxuser.com/V6/showthread.php?t=186726

I'm not disagreeing with you at all that some methods are better than others.

[Ian-T]

I can not think of a digital camera (high or low end) where I did not see aliasing. Moire is one thing....but I've seen aliasing from just about all of them (to some degree).

[Halsu]

All digital cameras create aliasing, because it is a square grid trying to render curved shapes.

The general use of the word aliasing refers to problems where the edge is not smooth enough - where a sawtooth edge, crawling or moiré can be seen at a normal viewing distance.

See i.e. these:
http://en.wikipedia.org/wiki/Aliasing
http://en.wikipedia.org/wiki/Moir%C3%A9


As for the Red not suffering from moire when down rez'd, well, people using Red cameras are talking about it.

http://forums.creativecow.net/thread/242/2516


I'll just quote myself:

"Downrezzing will not produce moiré unless done badly. The proper procedure may i.e. require blurring the image slightly before scaling etc. Just scaling the image down with your editing software is often not good enough..."

With proper treatment of the footage, small detail will be averaged together - i.e. in a resolution chart, when detail gets too small to be properly shown at target resolution, we see even gray instead of a moiré pattern.


The old fashioned way was to uprez an odd amount like 33%, then blur, then down rez. The blur creates some anti aliasing. Then sharpen a bit.

Here's an example of the same idea. Nothing is perfect, but it can save a shot.

http://www.dvxuser.com/V6/showthread.php?t=186726

Slight blur will help with aliasing and get rid of the moiré - the catch is, you need to blur the image before sensor, like in your latter example. There's simply no method that will cure moiré once it's introduced. And in the current DSLR's it's introduced in-camera (unless your image is out of focus) - there's nothing one can do in post to fix it.

[Halsu]

I can not think of a digital camera (high or low end) where I did not see aliasing. Moire is one thing....but I've seen aliasing from just about all of them (to some degree).

Sure - most OLP filters are tuned for a compromise. To get an aliasing-free image, it would need to be pretty darn soft, which is even less desirable than aliasing. This said, with current video cameras, aliasing is very rarely a problem.

A high resolution image scaled to lower resolution can be tweaked to maximize both - you can thoroughly control the aliasing / sharpness trade-in balance if you know what you're doing.

[Ian-T]

I don't understand why they don't just let us manipulate the native (large sized) image in post. I mean if for example a cam's native sensor is 5120x3000...let us save it at that size. I'm sure it could be sized down better in post or once captured. I wouldn't mind the extra step if it meant obtaining the best IQ. Seeing that current DSLRs are displaying higher data rates than ever before I don't see any "grave" issues with compression. I'll accept that compromise.

The best scenario would be RAW....I know....but for now ......

[Marshallator]

Current HDSLRs probably couldn't handle the processing of the native res images.

[Ian-T]

true that

[Duke]

Current HDSLRs probably couldn't handle the processing of the native res images.

Actually if its raw there is very little processing to do, just recording and if you choose it, in camera compressing.

(I'll preface this remark by saying its doesn't so don't bother arguing about it, but...) IF the 7D could record 1080 raw video at 24fps the data rate would be less than the 5120x3000 that its recording at 8fps. See below. Its not the data rate that's the hold up. Even 30 fps would probably fit.

1920x1080 = 2,073,600 x 24fps = 49,766,400 x 24bit = 1,194,393,600
5120x3000 = 15,360,000 x 8fps = 122,880,000 x 16bit = 1,955,080,000

Maybe the 7D lacks sensor speed. Could be a bunch of things.

[Ian-T]

The answer may be Wavelet based internal recoding like Cineform or JPEG2000. I would think it's much more difficult for these cams to process H.264 than it would be.....just about any other codec.

[Duke]

Yes, I think a wavelet codec would do it.

I also wanted to give a link to Barry's dynamic range tests. The 7D came in at about 7.5-8 stops which is pretty good.

http://www.dvxuser.com/V6/showthread.php?t=186746

[Halsu]

You are pretty much describing the Red cameras (both current and future ones). They save at the native sensor resolution, using wavelet compressed RAW.

[Halsu]

IF the 7D could record 1080 raw video at 24fps the data rate would be less than the 5120x3000 that its recording at 8fps. See below. Its not the data rate that's the hold up. Even 30 fps would probably fit.

1920x1080 = 2,073,600 x 24fps = 49,766,400 x 24bit = 1,194,393,600
5120x3000 = 15,360,000 x 8fps = 122,880,000 x 16bit = 1,955,080,000

Maybe the 7D lacks sensor speed. Could be a bunch of things.

Of course, to get that 1920*1080 raw image, the camera would need to use a cropped sensor (raw isn't scaled), probably the area would not that much larger than HV20... so a lot of the DOF benefits of using 7D would be gone.

I'd say 7D lacks both sensor speed and processing power for full sensor capture, it appears by a factor of 3X or so based on current 8 fps maximum and 24 fps target rate... but my guess is stuff like that will be available from Canon etc. a few years from now, according to moore's law.

[Duke]

Who said a full sensor capture. The 7D has twice as many pixels as the Red and even more than the Scarlet. Scanning every other pixel would speed it up and reduce the data stream.

Even then, what I said was 1080p. Notice that the red one doesn't scan full sensor at 2k or 1080p, so yes either skipping pixels or a cropped sensor.

The red camera is nice, but there's nothing magical about it. If Red can buy the sensors from a manufacturer, then so can Canon if they don't want to make them.

Its a silly argument though, to compare and $1,700 complete set up with a $34,000 complete set up. Talk to me again in a year and a half when we can both buy a Scarlet.



Of course, to get that 1920*1080 raw image, the camera would need to use a cropped sensor (raw isn't scaled), probably the area would not that much larger than HV20... so a lot of the DOF benefits of using 7D would be gone.

I'd say 7D lacks both sensor speed and processing power for full sensor capture, it appears by a factor of 3X or so based on current 8 fps maximum and 24 fps target rate... but my guess is stuff like that will be available from Canon etc. a few years from now, according to moore's law.

[Halsu]

Who said a full sensor capture.

I thought that's exactly what we were talking about?


The 7D has twice as many pixels as the Red and even more than the Scarlet. Scanning every other pixel would speed it up and reduce the data stream.

But that's what it's doing now: skipping pixels... that's what causes the horrid aliasing problems.


Even then, what I said was 1080p. Notice that the red one doesn't scan full sensor at 2k or 1080p, so yes either skipping pixels or a cropped sensor.

Yep - cropped sensor could work, but it would unsurprisingly also mean a crop factor of roughly 3X. That's about the same as 2/3" sensors - the resulting image would be similar to the cheap scarlets, though with less resolution.


The red camera is nice, but there's nothing magical about it. If Red can buy the sensors from a manufacturer, then so can Canon if they don't want to make them.

Well, Red must have done something right as no-one else can come even close, there's no other cameras of similar quality available anywhere near the same price point of their two year old model. In the future, that will probably change of course...


Its a silly argument though, to compare and $1,700 complete set up with a $34,000 complete set up. Talk to me again in a year and a half when we can both buy a Scarlet.

Half an year is probably closer...

[Ian-T]

Half an year is probably closer...You said that half a year ago....lol:hv20-smilie79:

[Halsu]

You said that half a year ago....lol:hv20-smilie79:

...and i still think i was closer to truth then than Duke is now ;-)

In other words, i think he's more overly pessimistic than i'm overly positive.

[Erik Bien]

Notice that the red one doesn't scan full sensor at 2k or 1080p, so yes either skipping pixels or a cropped sensor.

Among its various recording modes, RED One offers both of the "good" options:

2K Cropped — the camera crops a 'window' (just a smidge bigger than 1080p) out of the full sensor. The benefit is faster frame rates (up to 120fps) and compatibility with HD and 16mm lenses. The downside is decreased sharpness from the '4K strength' OLPF.
Quad HD (or whatever they're calling it). This records an image exactly twice as high and twice as wide as 1080p. The benefit is faster and better downsampling (not realtime, granted, but in post), since each 'finished' pixel uses data from a complete GRGB Bayer grid of photosites.


Skipping pixels not only reduces the resolution of the sensor, but also the fill factor (http://www.dpreview.com/learn/?/key=fill_factor): with all the engineering effort that's been dedicated to increasing fill factor in optical sensor design (microlenses, etc.), simply ignoring the data from half the photosites seems an inelegant solution to the problem of slow read-reset times.




The red camera is nice, but there's nothing magical about it. If Red can buy the sensors from a manufacturer, then so can Canon if they don't want to make them.

It's my understanding that, while the Mysterium sensor is most likely manufactured by an established chip fab, the design is proprietary to RED (http://www.reduser.net/forum/showpost.php?p=225355&postcount=6).

[Duke]

Its a silly argument though, to compare and $1,700 complete set up with a $34,000 complete set up. Talk to me again in a year and a half when we can both buy a Scarlet.


...and i still think i was closer to truth then than Duke is now ;-)
In other words, i think he's more overly pessimistic than i'm overly positive.

Oh, I think in 6-9 months some prototype beta models will be in the hands of those that are more than likely to give good reviews. They said Epic comes first.

And I think 12 months from now current Red owners will be able to get a Scarlet if they want a fixed lens.

I think 18 months or more before the general public can get them. I may be all wrong, but the Red company doesn't exactly have a reputation for delivering early, and they said they would beta test this time. I think that's being realistic, not pessimistic.

If they come earlier, great. :hv20-smilie77:

[Koppel]

I also think that Scarlet may be available for the public in 18 months.
So summer 2011 maybe... by this time a lot of other companies can come out with interesting gear as well.

[Ian-T]

Waiting for it would be like wasting your life. There is so much other things one could do between now and then. For those interested in Scarlet we all know that Epic will come out first. SO I think Duke's guess may be realistic.

[Halsu]

Oh, I think in 6-9 months some prototype beta models will be in the hands of those that are more than likely to give good reviews. They said Epic comes first.

My guess: less than two months.


And I think 12 months from now current Red owners will be able to get a Scarlet if they want a fixed lens.

My guess: 3-5 months.


I think 18 months or more before the general public can get them.

My guess: 5-9 months.


I may be all wrong, but the Red company doesn't exactly have a reputation for delivering early, and they said they would beta test this time. I think that's being realistic, not pessimistic.[quote]

It'll be interesting to see how it turns out... but i'll eat my imaginary hat if it takes as long as you suggest ;-)

[quote]If they come earlier, great. :hv20-smilie77:

...and if your estimates end up being right, i'm not too worried about it either - it's just another line of tools after all, albeit a cool one.

[Duke]

When we can get them in our hot little hands, then we can see. I'm hoping its tres cool or there will be a lot of disappointed fanbois. :hv20-smilie08:

[-Rogue5-]

I kind of got lazy and stopped reading half-way through the thread.

I was under the impression that the Canon's upcoming DigicV (note that the 1DmkIV and 7D use Dual DigicIVs and the 5DmkII use a single DigicIV) is to be Canon's first DSLR processor that is being designed from the ground up for both photo and video, with the intention of being able to do a full sensor scale down to 1080p with no pixel binning or skipping. So we may see it with the 1DsIV if we're lucky, but most likely the next wave of Canon DSLRs. I'm saying this is not so much as rumor but as logical evolution of Canon's HDSLRs (I have no inside info or anything).

When that happens, the *only* difference between a RED image and Canon HDSLR image will be that one is redRAW and the other will most likely be compressed (h264, hopefully better.) That's still a big difference, but not a $17,500 vs $1700 difference (unless the codec of the Canon's is actually pure crap.)

I've been testing all these cameras lately and one thing I've learned quite clearly is just how much of a liar vimeo is. I mean I took into account that its streamability and compression would weaken the image quite a bit, but I didn't realize that it actually hides flaws while still retaining sharpness. Even when you download videos from their site, it's been converted at least 3 times (raw - to editable format, editable to vimeo accepted format, and then vimeo's own encoding, which ends up being the downloadable one as well.)

So when you see intercut footage between different cameras (e.g. R1, 7D and 5D *cough*rodneycharters*cough*), even if it's a 1080p download, chances are the raw image was significantly different between the intercut cameras. I know that's to be expected, but I think most people still underestimate just how much of a difference there is. Same goes with low-light tests. The images are very useable and brighter and cleaner than any other cameras out there, but there still pretty friggen dirty (as in, not cinema-ready.) The more you clean them up, the more resolution you lose (which wouldn't be a big deal when you start at 4K res, but when you're just barely making 1080p, the hit can be quite substantial.)

-Rogue5-

[koolpenguin89]

The images are very useable and brighter and cleaner than any other cameras out there, but there still pretty friggen dirty (as in, not cinema-ready.)

I would like to take a pole of all of the people who bought a 5DMKII or 7D for video use, and see what percentage of those people have their products displayed in cinemas. Im willing to guess that the percentage is very, very low (as in, less than 1%).

Dylan

[-Rogue5-]

I would like to take a pole of all of the people who bought a 5DMKII or 7D for video use, and see what percentage of those people have their products displayed in cinemas. Im willing to guess that the percentage is very, very low (as in, less than 1%).

Dylan

heh, touche.

I guess what I should have said was, whereas you can take really dirty 4k RED footage with lots of high ISO noise, put it through a bunch of noise reduction and filtering, cutting down the resolution but giving you a nice clean image in the process, you can't do the same with HDSLRs. I find that with 5DmkII and 7D footage, any attempt to remove the noise usually results in your resolution dropping down to SD style (which isn't necessarily good for blowing up or even showing on a big HDtv screen).

Furthermore, I was more so pointing a finger at vimeo, which has this magic ability to make dirty footage of Canon HDSLRs (ISO3200 and 6400), look surprisingly clean & crisp, when in reality the raw footage is actually quite messy. ALSO, with RED footage, I'm finding that even with the noise, its such high resolution that it looks more like grain and retains its overall sharpness, but with 5D/7D, the noise actually starts to deteriorate the image quality/resolution.

I know it's not RED, but I was watching that show "The Hills" (don't ask) and a lot of it is shot in night clubs and even in HD it looks pretty clean. I don't know if 5D or 7D footage would look that clean unless it was broadcast in SD.

It's almost like Canon's footage brighter at the cost of resolution while (literally and figuratively) RED has resolution to spare (on noise reduction).

-Rogue5-