MattH

Nice demonstration. 

The making of shot on 5d3 raw looked pretty good as well.

Did you get a chance to do any comparisons between them?

[Deleted by EOSHD - off topic, sorry. Please read my posts above.]

I hope it's not bothersome as I already asked it before but here we go again, hopefully I'll get a reply this time.

 

In theory, what should be the differences for the GH4 in terms of IQ between:

 

1) GH4's 1080p (@100mbps IPB or @200mbps All-I)

2) 1080p via external recorder ProRes 10-bit 4:2:2

3) 1080p resized from GH4's 4K footage (post)

4) 1080p crop from GH4's 4K footage (post)

 

If there is, which should have the best sharpness, DR, less artifacts, less moiré, etc?

1. Can't say for sure but this will probably be fairly similar to what the gh3 already does, perhaps a little better

2. Much more latitude for grading due to the 10 bit  (4 times as many levels to work with), twice as much chroma resolution for smoother colour gradients and less aliasing round edges of high colour difference like red against white.

3 (assuming from in camera 8 bit 4k) same grading latitude as internal 1080p.  Will definitely look sharper displayed at 4k vs up-scaled 1080. Will definitely look sharper if you have to re-frame or zoom in compared to 1080 zoomed in. Will most probably look sharper when downscaled to 1080 than 1080 due to the oversampling, even if the bit rate is lower.    No all I-frame in internal 4k so complex movements may be rendered worse in the compression process than all I-frame 1080p.

4. will have the least detail of all as you'd be cropping out 75% of the 100bps information.  Would probably be acceptable in certain shots but you wouldn't want to make a habit of it.  Better to use a longer lens or use the in camera 1080 crop.

All in all I think the right setting would depend on what you wanted from that particular shot.  If you wanted more spacial resolution you would chose 4k. if you wanted a wider field of view, the narrowest field of view, slow motion options or more temporal resolution you would go for 1080.

I don't know, I just worked it out in theory.  Perhaps it does need a particular method, but it should be possible.  Hopefully somebody else will know.

I've read a few times now people say on the internet that 4k 4:2:0 when scaled down to 1080 becomes 4:2:2.

Can somebody shed some light into this? Is this true or just nonsense?

Thanks!

I was going to just confirm this, but looking into it, Its actually better than that:  Its actually 4:4:4

With 1080p 4:2:2 you have 1 chroma sample for every 2 output pixels for a total of  1,036,800 chroma samples.

with 4:2:0 you have 1 chroma sample for every 4 output pixels for half that amount (518,400) .

But since the number of output pixels in quadHD 4K is four times as much as 1080, even with 4:2:0 you still have 2,073,600 chroma samples.

That's a chroma sample for every output pixel when you downscale to 1080.

Very interesting.. :)

The external add-on isn't a "4K add-on" it is a HD-SDI and XLR add-on, you don't need the external add-on or a external recorder for 4K on the GH4 it does it internally. You know that right?!

I think people (myself included) are confused about whether you need the xlr plinth to get 10bit 422 4k output specifically.  I got the impression that was the case.   I assume we can get 10bit 422 1080p output from the gh4 itself via the hdmi output.  If we can get 10bit 422 4k output direct from the camera as well that will simplify things a lot.   

To be exact I think it's 2.17x in 4096 and 2.33x in 3840.

 

In order words the 4096 x 2160 DCI 4K mode is 17% smaller than the sensor area used in 16:9 1080p mode.

I worked the crop factors out at 2.5 for quad hd and 2.3 for cinema 4k.

The maths:

width of (the imaging area of) the sensor (17.3) divided by the horizontal sensor resolution (4608) times the crop horizontal resolution (3840,4096) gives the horizontal width of the crop which is 14.41mm and 15.57mm respectively.

Divide 36 by these widths to get crop factors of 2.498 and 2.312.

(for lens image circle coverage purposes they have diagonals of 16.53mm and 17.55mm [you can google pythagoras if you want the maths on that ])

One question I have is does it have an anti aliasing filter.  stills cams usually do (besides from the resent trend).  If it does, this will be a big advantage over the bmpc which wont have one.  of course the filter makes things softer but when you downscale to 1080 its crisp but also clean.

Regardless of the codec specs i think it will come down to the overall image character in terms of things like smooth highlights and colour etc that will prove whether this camera wins compared to its rivals.

Not sure if I'm interpreting this post correctly but you can slice and dice this list up a 100 ways depending on the particular job. For instance you have the 5d higher than an F3 with slog. On a doc job, under various scenarios I feel you'll get a better image out of the f3 than the 5d. A more controlled situation you maybe right. Plus (with the odyssey) I like the fs700 over the red one MX any day. 240 fps 2k RAW, 4k compressed (soon). 12 bit 444 1080 off the 4k signal. Ive seen it and it lokks like a differnt camera, great image. For narrative, commercial it's a different story, maybe you should've put a 35mm film camera before the Alexa if only image quality was your focus? I enjoyed the read but think choosing the right camera for a job boils down to more than just the image.

It doesn't matter as no-one is beholden to the list.  I think it just puts things in the correct ball park.  Useful for beginners.

Not true.  Not true at all.  I know people who have spent more than this, either individually on a RED or a group of local shooters pooling their cash to buy one and share.  This is likely to give them all serious buyer's remorse.

 

Rednecks spend more than this on fishing boats, hunting cabins and off-road vehicles.  Yuppies spend several times over this price for a midlife crisis sports car, a sail boat, a motor home or on vacation.  I have friends with much more than this invested in their gun collection.  A buddy of mine has spent far more than this building an airplane kit in his garage.

 

What's perhaps ironic is this camera at this price might be privately owned by individuals who are non-photography professionals or otherwise not in full-time employment as shooters.  Folks with six figure incomes, access to credit and looking for another expensive hobby, this one with the benefit that their purchase could eventually pay for itself by renting it out.

You live in a very different world to me, my friend.

when you were talking about noise you said:

"it has a slightly gritty texture to the shadows but that’s something you can’t crush back a bit in the grade to hide."

Did you mean   “but that’s not something you can’t crush back a bit in the grade to hide.†?

it looks like you changed from "that's something you can crush" but forgot the "not".

I've just done a very crude test using the APS-c crop mode on my A7R.

 

Full Frame:-

https://dl-web.dropbox.com/get/FF.jpg?w=AAC7BwWnC-dDmV8Wabg5d50h4rcCnqQcK82p5zzyZFocww

 

APS-C + Speed booster

https://dl-web.dropbox.com/get/SB.jpg?w=AACS4qZtmb-7CXtvNNwQgh8y0-JIuzzvTDfEd2-PNKcGgg

 

point of focus was on the eye balls of the squirrel.  the lens was a zeiss planaer at f2.8.  I downsized the large 36mpx image and overlayed it over the aps-c shot to even the odds resolution wise (the full frame shot was 36mpx and the aps-c shot was 16mpx..

 

notice how the defocus on the lenses in the background is the same, yet more of the squirrel's face and shoulders are in focus on the full frame shot.   this should be less obvious as you move further away from the subject, however to me I seem to see it becomes even more obvious on medium/wide shots where the distance between the camera and the subject is greater.

I just signed up for dropbox but the links are still giving me a Dropbox 403 error.  It it something i'm doing wrong?

Thanks, very clear explanation - but I'm struggling with the last paragraph. I'm probably being really slow here, but why will the noise be the same? When you boost ISO doesn't the signal-to-noise ratio deteriorate?

 

Of course there is also the 'real world' issue that the larger sensors we're talking about are generally on hybrid cameras that often downsample in ways that don't make all of the surface area count toward signal-to-noise. Whereas the Blackmagic cameras are native video resolution, so have larger pixels with full readout (i.e. all surface area does count). Correct?

Its general equivalence theory.  Larger sensors generally have lower noise at higher ISO.  Assuming the sensors are of the same generation.  Its what makes a full frame camera better than a point and shoot.   The speed booster effectively makes your sensor bigger so you are now equal with the larger sensor but not better.  The speedbooster doesn't give you a free meal.

And yes you are right, black magic use all of their sensors so that is a noise advantage compared to an a camera that doesn't.  Magic lantern raw on the 5diii for example probably only uses 1/9th of its pixels.

Shouldn't this work on the GH2?

 

Correct me if I'm wrong, but if it's only the physical shutter getting in the way, then the GH2 should be fine right? As in, the default state of the shutter on the GH2 (Even when powered off) is open. So video would be fine. Just don't take photos with it?

Id imagine that this new speed booster is optimised to put an apsc sized image onto a 2.3* crop sensor, and so wouldn't cover the sensor of the GH2 and you would get a black ring.  Whereas the original micro 4/3rds speed booster is designed to put a roughly apsh sized image circle on to a regular micro4/3rds sensor.   I cant be sure, but that's what I would guess.

Would you mind explaining this a bit further please, or putting it a different way? I can't quite get my head around what you're saying …  :unsure:

 

EDIT: Don't worry, I just got it ...

 

No, actually I didn't, lol  :lol:

Ok. Let’s say we have a Camera A with a large sensor and camera B with a smaller sensor. And we have two identical lenses to put on them.

For simplicity we will say that the area of the Sensor A is twice as big as Sensor B.   And that when you put a speed booster on camera B  the same image that the lens delivers to Sensor A is compressed to fit on sensor B.  So we have the same angle of view and the same depth of field.

Let’s say that sensor A’s area is 100mm squared and sensor B’s area is 50mm squared.

Let’s say for simplicity that the lens is letting in 1 photon of light per square millimetre every 1/50^th of a second.  So sensor A has 100 photons of signal every frame.

If we imagine camera B without the speed booster, it isn’t getting light from all the lens. The lens is still letting in 1 photon per square millimetre every frame.  So camera B without the speed booster has 50 photons of signal per frame.

Exposure or brightness is light per surface area.  So even though sensor A has more photons per frame, When you divide the number of photons by the area both cameras in this case have the same: 1 photon per square millimetre so they have the same brightness.

Now if we put the speed booster on camera B the full image from the lens: All 100 photons is compressed onto sensor B.  So we now have 100 photons and 50 square millimetres. So we have 2 photons per square millimetre.  The image is therefore twice as bright in camera B.

But both camera A and B are getting 100 photons of signal per frame.  The same signal to noise ratio.   So even though you will have to boost the iso/gain of camera A to match the brightness of camera B, the noise at that higher iso will be the same as the noise of camera B at the lower Iso.

Would you mind explaining this a bit further please, or putting it a different way? I can't quite get my head around what you're saying …

 

EDIT: Don't worry, I just got it ...

OK. cool. Although I think I have another way to explain it if you need.

Unfortunately the comment is slightly wrong.  the effect is actually changing the lens focal length.  the ratio between defocused infinity and an 'in focus' subject in the foreground will indeed be the same as on a bigger sensor without the speed booster, but the actual look can be quite different.  

 

assuming you took a 50mm f1.2 lens on full frame, and put it against a 80mm f2.8 lens on medium format, the ratio of 'maximum defocus' to 'in focus' and the fov would be similar, but the actual rendered image would be drastically different in terms of depth of field.  the medium format image will have more ping to the in focus subjects.  assuming the point of focus were the eyes of a model, the medium format setup would capture a deeper depth of field while still creating as strong maximum defocus.  Therfore the medium format image will render a greater amount of the face in focus than the faster lens on a smaller sensor.

 

the same differences will be seen when comparing a boosted pocket vs an aps-c sensor.  both will have the same ratio of defocus to focus, and the same fov.  but the aps-c will render its depth of field in a more sectioned off and 3 dimensional fashion.  the pocket will be over a stop brighter though.

There are two questions at hand here.

1. does this (the effect you mention) happen

2. Does this happen with the speed booster.

1.  I have always had a feeling that this happens.  Smaller sensors with faster lenses never seem to give as strong a subject isolation.

However, I have never seen someone prove this, either with a demonstration or with scientific theory.  If you have more info on this I would love to hear it to confirm my suspicions.  I would also love to see you try and present your case in the dpreview open talk forum.  Particularly with a provocative thread title. There are some hardcore equivalency crusaders there, and you would really get their hackles up

2.  I don't actually think it is happening with the speed booster.  It isn't actually changing the focal length at all.  The booster cant change the properties of the lens. Its just taking the image that comes out of the back of the lens and squeezing it into a narrower image circle.  besides any distortion or refraction or whatever that the booster ads, the image will look the same apart from minor variations in the crop.

(incidentally, for the people saying the speed booster increases iso performance.  It depends.  increasing it compared to what?

Compared to using the same lens on the BMCC without the speed booster? Yes.

Compared to using the same lens on a native apsc sensor camera? No.  The image will be brighter at a lower ISO, but the noise will be the same as the apsc camera at a greater ISO.   The signal to noise ratio is the same, because the signal is the same. It cant get any extra signal from somewhere.)

Thanks Andrew, this clarifies everything I was confused about, I think

 

Though I am still unsure about the brightness. I get that, as you say, F0.74 remains so regardless of crop factor. But then a smaller sensor always gathers less light than a larger sensor doesn't it? So doesn't that mean that, all other things being equal, an F0.74 lens on a Super 16mm sensor will not result in as-bright an image as a 0.74 lens on a Super 35 sensor? And won't that difference in brightness be mathematically relative to the crop factor (i.e. the same multiplication the other things)? Again, I hope I've made this clear … it's perfectly possible I'm missing something.

 

Amazing low-light footage btw - certainly gives a different angle on "the camera discussion that shall not be named"

An F0.74 lens will give the same exposure on any size of sensor.  The exposure is unconnected to the sensor size.  A bigger sensor collects more light energy (more photons or whatever you want to call it) for a given f number but the photons are spread over a larger area so it isn't brighter.   The speed booster concentrates this light energy into a smaller area like a magnifying glass burning an ant, so the exposure is increased but the noise will be the same if both sensors are equally good.

The easiest way to think of the speed booster is by acting like it modifies the sensor.  Giving the BMCC an APSC crop factor. Giving the same depth of field and noise as an APSC camera, but giving a brighter image for a given iso.

Great to have new options, but the inability to use on regular m4/3 cameras seems a bit of a disadvantage to me.  It may not be as future proof because a new camera - even from black magic - may have a different crop factor or a larger shutter or built in NDs.   

Also one thing to bear in mind is that - even with these current cameras - because the rear element is close to the sensor a mosaic anti aliasing filter may not be able to fit behind it.

For those who already have the original on the cinema camera I think its good to understand exactly what the difference will be.  The native crop of the bmcc is 2.2777  (36mm/15.81mm)  (I don't know where 2.39 comes from. perhaps it doesn't account for aspect ratio).

So with the current speed booster you have an effective 1.617 crop (which is pretty much equal to canon apsc) and with the new one you will be getting a 1.457 crop.  So its a noticeable difference but nothing earth shattering.  Although 1.457 is wider than the 1.5 crop that apsc lenses are designed to cover, I think most lenses will have enough leeway in their image circle for this not to be an issue.

Well done sir! Finally. You've started to apply numerical ratings! Don't be too generous though. 7/10 might be enough for Sony to call it a Big Win!
Further. "A small rant" should be a pinned block of bold text in your reviews banner, to remind the makers that this is our cardinal lamentation!
 

I agree with this.  If we are talking primarily about video, and 10/10 is alexa like quality in a small form factor, is the RX10 really 7/10? With limited dynamic range, aliasing and compression mud i'd say more like 5/10.  An A- for effort with the hardware features but D for attainment with the lousy processing and codec. And it's not exactly cheap either.

Weird for you to make the under a rock comment since this is the first time you have ever mentioned it in an article on your site as far as I can remember.  But it's an interesting potential future development for sure.  I'm guessing this will more useful for content delivery than acquisition due to the high processing required.  And i'm actually hopping that SD and CF card speeds and capacity will increase quickly enough that it will render it pointless for high quality acquisition anyway.  

GM1 video quality.

With the Panasonic GM1 having the same sensor and processor as the GX7 I would assume that they would have identical video image quality at the same settings.  The GM1 doesn’t have 60 fps, which is presumably down to heat.  But at 24 frames a second it should be the same.

Video samples from the GX7 that I have seen seem to be pretty much on par with the GH3,  however samples from the GM1 on YouTube seem to be lower in quality.  Less detail and obvious aliasing on diagonal lines.

Is this due to the settings used by the testers, or how they have uploaded the video?

Or could it be due to the automatic corrections being done for the GM1kit lens?

Or could Panasonic have artificially lowered the quality for some reason?

Here are some examples (obviously view natively at 1080p for maximum possible quality)

http://www.youtube.com/watch?v=j49oMnhX21U

http://www.youtube.com/watch?v=Ati36TwEXJI

The next one looks particularly mushy.  I’m hoping it’s because it hasn’t been focused properly.  Front focused with a high f number maybe?

http://www.youtube.com/watch?v=lcnUfDY6cCc

If you look at the test for the GX7 from the same people it looks remarkably sharper:

http://www.youtube.com/watch?v=t_g-gVfuMpQ

In the dynamic range test the 5d3s shadow areas are noticeably brighter than the pocket cams.  So I think 2 stops difference is exaggerating it a bit.  And the highlights are only a little bit brighter so I would actually say they are pretty equal for dynamic range.

I noticed a bit of rainbow moire on the side of the camera on the pocket low light test. Not bad, but its there.

Did you sell your cinema camera? It would be grate to see it included in comparisons.

They are just what I expected. Consumer compact still cam video quality. Because that's what it is really  I don't know how the DVXuser guys could be so optimistic.  And I can't believe they would see this footage and still maintain their optimism. This footage is in no way suitable for broadcast.

Wow, it's like night and day. Most of the people over on dvxuser are mightily impressed by this camera and are singing it's praises and over here most of you seem to think it's crap. What gives?

Well, I said most people. The only really negative view was by one person, who made a number of posts. Someone mentioned the lack of 30p but said that was their only complaint. The overwhelming view over there is positive, very positive.

I was going to say maybe they were Sony fanboys not having an objective view.  But after reading the thread I actually don't think that is the case.  I think the people there are videographers in the traditional pre-dslr sense of the word.   They are seeing this cam as a convenient all in one workhorse for ok videoey looking video.   Whereas people here are more interested in ultimate image quality.  We judge dynamic range, detail and tonality to the same standard a still photographer does with stills.  It doesn't matter to us if a camera is twice as convenient: If it doesn't represent a significant increase in image quality for its price point then we aren't interested.

Audio duly muted!