AG-DVX200 4K Large-Sensor Handheld Camcorder

[joe 1008]

On the other hand, I believe that it's a shame that Panasonic shipped the DVX200 with some evident problems on colors and noise, leaving the hard work to its customers.

I believe something went profoundly wrong here and I see some Panasonic reps hiding their heads in fear. Panasonic has proven that they handle 4/3" sensors. They handle the whole technology. There's nothing exceptional about the DVX200 but that it's the first of it's kind. A state of the art machine, a 2/3" camera in the size and form of the famous DVX100. But for some reason they leapt too fast into the public with this product. Maybe because of the pressure of other camera producers releasing competing products. But it happened and now we have to give time to this to settle. Of course it's a shame for those who already bought this camera. But I'm convinced that all will turn out right and that Panasonic will provide the due service to the early buyers and finally offer a product that really is capable to become the new DVX.

 

[Alpha Beta Omega]

I see the noise at 100% on an iMac. But it's not as bad as pancam's footage in which the noise on the jacket at 100% is just a distracting swarm.

The 300% part it looks shitty on the DVX200 and with false contouring reminding me of tellys with SVM engaged. At 100% it's less evident but it has no smoothness as the G70 have.

Having tried all parameters on the AF100, suspecting the DVX200 is very similar, it's just futile to try and get the good stuff out of these two cameras AF100/DVX200. There's just bad electronics in the works and mind blowing that you have to struggle with scene files and what not to try and get something good out of it. Especially when there's so many other cameras that just look good without any trickery.

 

[Htech]

The problem is black colour of the jacket. This noise apear in dark colours, shadows. If you cut half of that studio light the noise will be worse. This is main problem for run and gun in natural/dim/indoor/given light! If we do something wrong why nobody can give us a pertinent solution? Neither Panasonic.

 

[Barry_Green]

Been plinking around today trying to get something that reasonably matches the GH4's "Natural" profile. Having a GH4 on hand should make it much easier to match them; I was handicapped in trying to do it last time.

I haven't yet gotten the gamma to give me exactly the same on both, but the color seems reasonably close. Certainly enough that if someone likes the look of a GH4 on Natural, they'd perhaps like to give this scene file a go and see if they get what they like.

View attachment 107817View attachment 107818

The scene file is available for download at http://www.fiftv.com/DVX200/Scene_Files/GH4-Natural/SCENE1.TXT.
It is SCENE 1 in that file.

 

[Barry_Green]

But it's not as bad as pancam's footage in which the noise on the jacket at 100% is just a distracting swarm.

And yet, on my system, or on the Odyssey, there's no swarm whatsoever, and most definitely not distracting. Not until you blow it up to 400%...

 

[Xoomtell]

I believe something went profoundly wrong here and I see some Panasonic reps hiding their heads in fear. Panasonic has proven that they handle 4/3" sensors. They handle the whole technology. There's nothing exceptional about the DVX200 but that it's the first of it's kind. A state of the art machine, a 2/3" camera in the size and form of the famous DVX100. But for some reason they leapt too fast into the public with this product. Maybe because of the pressure of other camera producers releasing competing products. But it happened and now we have to give time to this to settle. Of course it's a shame for those who already bought this camera. But I'm convinced that all will turn out right and that Panasonic will provide the due service to the early buyers and finally offer a product that really is capable to become the new DVX.

Let us hope Panasonic finds out how to resolve these issues very soon.
Sony did aparently get a very much better noisereduction just by a simple firmware update for the FS5 recently, so why should that not be possible for Panasonic as well. They have at least tried to do something with the latest firmwate, and some saw a difference, others did not. I think there will be new updates very soon.

https://vimeo.com/116047140

 

[Xoomtell]

Been plinking around today trying to get something that reasonably matches the GH4's "Natural" profile. Having a GH4 on hand should make it much easier to match them; I was handicapped in trying to do it last time.

I haven't yet gotten the gamma to give me exactly the same on both, but the color seems reasonably close. Certainly enough that if someone likes the look of a GH4 on Natural, they'd perhaps like to give this scene file a go and see if they get what they like.

View attachment 107817View attachment 107818

The scene file is available for download at http://www.fiftv.com/DVX200/Scene_Files/GH4-Natural/SCENE1.TXT.
It is SCENE 1 in that file.

testshots please, some one?

 

[bluesgeek]

And yet, on my system, or on the Odyssey, there's no swarm whatsoever, and most definitely not distracting. Not until you blow it up to 400%...

Barry, here's a screen shot from my iMac 17" at fullscreen. What appears a fabric texture is actually noise.

 

[harddrive]

....... please tell me how pixel area to sensitivity comparison computations are made? If there were 8 times more pixels would that be a 2-stop advantage, 4 times be a 1 stop advantage?

When I commented about (theoretical) sensitivity comparison computations, I was going simply on area of chip - ignoring pixel count. Rule of thumb is that every time the area of the chip doubles, the base sensitivity will increase by a stop. And as a first approximation, going 1/3" - 1/2" - 2/3" - 1" - 4/3 is a rough doubling of area each time. So for my quick and dirty calculation, the AX1 after cropping is approximately 1/3", the DVX200 is 4/3 - so approximately 16x the area. So I'd expect it to have a 4 stop ISO advantage - but half of that is negated by a f2.8 lens versus f1.6 of the AX1. So the expectation is the AX1 would be about 2 stops better in low light. And the video shown seems to roughly agree with that.

If the sensors are like for like apart from area (same pixel count), it follows that each photosite of the bigger chip must also then be the same proportion bigger. So for the same intensity of light falling on the chip (governed by f no), if the chip is 16x bigger it follows that each photosite will be 16x the area - so for a given exposure will collect 16x as many photons of light. So it should be obvious why the base ISO relationships increase with increasing chip area. Practically, it should be a little better than those estimates, as the bigger the photosites, the less area "wasted" (for light gathering purposes) by the band between photosites.

Increasing photosite count is interesting, but if you assume the same output resolution is the same in each case (say QFHD, or 3840x2160) and both cam A and cam B have the same area of sensor, but cam B has 4x the photosite count of A, then at first sight it may seem likely that cam A will be 2 stops higher ISO than cam B? (It will have a quarter the photosite count - but each photosite will have 4x the area.) But assuming all photosites are used, then there should be little difference between the overall results and that's down to the downconvert process that cam B will need effectively averaging photosite outputs. Rather crudely, think of it as 4 smaller photosites averaging to give the same as a single larger one. In other words, for all else equal (including output resolution) the base ISO is likely to be proportional to ONLY the area of the used part of the chip. The photosite count becomes irrelevant.

I've stressed "all else equal" in these (approximate) calculations - it's worth thinking about a few things which may NOT be equal, and which may interfere with even a rough calculation! First, are we talking about single or three chip? For a given size (say 2/3" camera) the use of 3 chips and beam splitter will inevitably give a better result than single chip, as no light gets lost via absorption in the Bayer filtration. Again as a rough approximation, I'd expect a 3 chip 2/3" camera to be about 1-1 1/2 stops better than an equivalent single chip 2/3".

Secondly, I previously said "assuming all photosites are used" and whilst that is nowadays increasingly the case, it hasn't always been so. Using a high photosite count sensor for video first meant simply omitting a proportion of the photosites - pixel-skipping. The technology simply couldn't cope with reading all the photosites of the sensor at video framerates. Initial attempts gave some pretty poor results (bad colour aliasing), but techniques improved and such as the AF100 apparently read a single 2x2 Bayer block out of each 4x4 block, and got R,G,B values directly for each block. But only making use of a quarter of the total photosites meant that for base ISO purposes such a camera gave the results expected from a (fully read) chip of a quarter the size - so in the case of the AF100, comparable with a single 2/3" chip or a 3 chip 1/2" design.

And of course, the AX1 is an example of not all the photosites being read in another way, in this case just reading the centre portion of a higher count chip - which is why I count the AX1 as roughly a 1/3" chip, rather than the 1/2.3" it physically is.

I stress that all this can only give an approximation to the first order, but it can nonetheless be useful in prediction.

 

[bluesgeek]

When I commented about (theoretical) sensitivity comparison computations, I was going simply on area of chip - ignoring pixel count. Rule of thumb is that every time the area of the chip doubles, the base sensitivity will increase by a stop. And as a first approximation, going 1/3" - 1/2" - 2/3" - 1" - 4/3 is a rough doubling of area each time. So for my quick and dirty calculation, the AX1 after cropping is approximately 1/3", the DVX200 is 4/3 - so approximately 16x the area. So I'd expect it to have a 4 stop ISO advantage - but half of that is negated by a f2.8 lens versus f1.6 of the AX1. So the expectation is the AX1 would be about 2 stops better in low light. And the video shown seems to roughly agree with that.

If the sensors are like for like apart from area (same pixel count), it follows that each photosite of the bigger chip must also then be the same proportion bigger. So for the same intensity of light falling on the chip (governed by f no), if the chip is 16x bigger it follows that each photosite will be 16x the area - so for a given exposure will collect 16x as many photons of light. So it should be obvious why the base ISO relationships increase with increasing chip area. Practically, it should be a little better than those estimates, as the bigger the photosites, the less area "wasted" (for light gathering purposes) by the band between photosites.

Increasing photosite count is interesting, but if you assume the same output resolution is the same in each case (say QFHD, or 3840x2160) and both cam A and cam B have the same area of sensor, but cam B has 4x the photosite count of A, then at first sight it may seem likely that cam A will be 2 stops higher ISO than cam B? (It will have a quarter the photosite count - but each photosite will have 4x the area.) But assuming all photosites are used, then there should be little difference between the overall results and that's down to the downconvert process that cam B will need effectively averaging photosite outputs. Rather crudely, think of it as 4 smaller photosites averaging to give the same as a single larger one. In other words, for all else equal (including output resolution) the base ISO is likely to be proportional to ONLY the area of the used part of the chip. The photosite count becomes irrelevant.

I've stressed "all else equal" in these (approximate) calculations - it's worth thinking about a few things which may NOT be equal, and which may interfere with even a rough calculation! First, are we talking about single or three chip? For a given size (say 2/3" camera) the use of 3 chips and beam splitter will inevitably give a better result than single chip, as no light gets lost via absorption in the Bayer filtration. Again as a rough approximation, I'd expect a 3 chip 2/3" camera to be about 1-1 1/2 stops better than an equivalent single chip 2/3".

Secondly, I previously said "assuming all photosites are used" and whilst that is nowadays increasingly the case, it hasn't always been so. Using a high photosite count sensor for video first meant simply omitting a proportion of the photosites - pixel-skipping. The technology simply couldn't cope with reading all the photosites of the sensor at video framerates. Initial attempts gave some pretty poor results (bad colour aliasing), but techniques improved and such as the AF100 apparently read a single 2x2 Bayer block out of each 4x4 block, and got R,G,B values directly for each block. But only making use of a quarter of the total photosites meant that for base ISO purposes such a camera gave the results expected from a (fully read) chip of a quarter the size - so in the case of the AF100, comparable with a single 2/3" chip or a 3 chip 1/2" design.

And of course, the AX1 is an example of not all the photosites being read in another way, in this case just reading the centre portion of a higher count chip - which is why I count the AX1 as roughly a 1/3" chip, rather than the 1/2.3" it physically is.

I stress that all this can only give an approximation to the first order, but it can nonetheless be useful in prediction.

One of those classic posts which I will read many times next summer... and I have already learned a great deal from it. Thank you very much for taking the time to provide this very useful information.

 

[harddrive]

Barry, here's a screen shot from my iMac 17" at fullscreen. What appears a fabric texture is actually noise.

My first thought is that it looks like a quantisation artefact, rather than normal noise. It's the sort of thing I'd expect where quantum values can only take a limited number of values - so say have to be either 0, 4, 7, 10 etc rather than being able to be any of a full range - 0,1,2,3,4,5,6 etc.

So you get a "clumping" of pixels of the same value which can make the effect more visible than pure random noise. So if you had 9 pixels, and noise meant their values came out as 4,6,7,5,4,4,7,6,5, that would be a lot less visible than a quantisation effect which gave something like 4,4,4,7,7,7,7,4,4.

 

[Greg Smith]

I've stressed "all else equal" in these (approximate) calculations - it's worth thinking about a few things which may NOT be equal, and which may interfere with even a rough calculation! First, are we talking about single or three chip? For a given size (say 2/3" camera) the use of 3 chips and beam splitter will inevitably give a better result than single chip, as no light gets lost via absorption in the Bayer filtration. Again as a rough approximation, I'd expect a 3 chip 2/3" camera to be about 1-1 1/2 stops better than an equivalent single chip 2/3".

I tend to visualize this in a slightly different way. 3-chip cameras also have the equivalent of color filters, in the form of dichroic mirrors, incorporated into the prism system, so they experience similar light loss as Bayer sensors, although it's done with selective reflection rather than absorption. Roughly 1/3 of the total light energy still gets to each chip. On the other hand, they have 3X the total chip area by definition, and the individual photosites on each chip can be larger, increasing sensitivity. I think your 1.5 stops overall advantage is probably about right -- this is due to my empirical observations as much as the math -- but it is due to different factors than the filters alone. Three chip systems also have an advantage in potential resolution and color bandwidth, of course.

- Greg

 

[harddrive]

Whilst agreeing with what he says, I'd probably put the general advice the other way round - take the coring level well down, then set detail level to suit.

I say po-ta-to, you say po-tah-to

Suppose I asked for that, but I still thinks there's a difference - you want to end up with very little coring - end of story - so IMO it makes sense to simply wind it well down first. Then you see how little overall detail you can get away with, but it does mean you're doing it with the coring out of the equation. Wind down detail and then think of coring and you could end up chasing your tail.

I think it's also worth saying that with adjustments like these it's possible to see them make a huge change on one type of shot, but any difference be barely visible on another. And I believe that's been a big trap in the past with coring being presented as a "noise reducer". It's been set on a non-critical scene with the eye primarily on the noise, then it's only later (typically with a face in shot) that the downside becomes obvious. With the danger it gets diagnosed as "just looks a bit soft", so up the overall detail, so up the noise goes, and.......

 

[harddrive]

3-chip cameras also have the equivalent of color filters, in the form of dichroic mirrors, incorporated into the prism system, so they experience similar light loss as Bayer sensors, although .......

Hmmm, not sure I really go along with that...... In a 3 chip camera the whole point is *NO* light gets "lost" (there's no absorption), it just gets differentially directed to the appropriate chip according to colour. So all the incident light energy can be made use of. (And is.) With a Bayer sensor, then for a red photosite all the green and blue light that falls on it gets absorbed, so can more truly said to be "lost".

Three chip systems also have an advantage in potential resolution and color bandwidth, of course.

With a caveat of "all else equal", then yes. I think you're taking as read the three individual chips have the same size and pixel count as the single Bayered sensor?

Obviously a single 3840x2160 Bayer sensor will have better resolution and colour bandwidth than 3x 1920x1080 chips.

 

[Bassman2003]

I just downloaded Pancam's studio footage and put it in Edius and viewed it on my 20" broadcast monitor in 1080p. From my view I only see a little movement in the jacket. Exactly what Barry is saying in that it is nowhere near the catastrophe that some are making it out to be. Could it be better at 0 db in a studio, yes. Is it horrible? no. Can the AC-160 do a lot better? I would want to see them side by side because what is considered noisy today was whistle clean a few years ago.

Now what is bothering me is the AF100 style semi posterized skin tones. It just looks over processed. Hopefully this can be improved.

 

[Greg Smith]

Hmmm, not sure I really go along with that...... In a 3 chip camera the whole point is *NO* light gets "lost" (there's no absorption), it just gets differentially directed to the appropriate chip according to colour. So all the incident light energy can be made use of. (And is.) With a Bayer sensor, then for a red photosite all the green and blue light that falls on it gets absorbed, so can more truly said to be "lost".

Viewed at the pixel level -- in a Bayer sensor, 2/3 of the light is absorbed by the filter over the photosite -- in a 3-chip prism camera, 2/3 of the light is reflected away to the other two chips. Same level of illumination at the pixel -- but there is three times as much total area in a 3-chip camera to sense it.

With a caveat of "all else equal", then yes. I think you're taking as read the three individual chips have the same size and pixel count as the single Bayered sensor?

Obviously a single 3840x2160 Bayer sensor will have better resolution and colour bandwidth than 3x 1920x1080 chips.

I was thinking of the case of equal chip size and resolution, e.g. one 2/3" 1920x1080 Bayer chip versus three 2/3" 1920x1080 chips and a prism. The latter would obviously be a larger and more expensive system, but would have approximately that 1.5 stop sensitivity advantage, as well as being able to be decoded to 4:4:4 color if the processing supported it. (Actually, it's remarkable to me just how cheap the smaller 3-chip cameras have become.)

A 3840x2160 Bayer sensor will just exactly match a 3x1920x1080 system in resolution in the red and blue channels. Beats it on decoded luminance resolution, however, but it's not a full two-to-one advantage. Most 3-chip cameras don't encode all the color information they would theoretically have available off the sensors.

Another "not all equal" factor is that it's more practical and economical to design fast lenses for smaller sensors. To bring this back on topic a little, I have both a PX270 and a DVX200. The PX has three chips at 1/16 of the area each compared to the DVX, but the lens is about 1.5 stops faster (varies a little through the zoom range). My (fairly repeatable) results are that the PX270 has just about exactly a one stop overall advantage over the DVX200 in low light given the same gain and high-sensitivity settings.

- Greg

 

[Greg Smith]

I have no trouble seeing the noise in the newscaster's jacket in Pancam's studio shots. It's pretty bad, actually. I'm using my 30" Dell computer monitor, which does, no question, lift the blacks a little bit compared to a calibrated video monitor. But it's still a worse performance than I've seen from other contemporary cameras under the same viewing conditions, and I can well imagine what he sees when he does a split screen comparison.

But I have footage from my own DVX200 that has a similar tonal range, including nearly identical black fabrics, and absolutely does NOT have this level of noise in the blacks and shadows. I don't know where it is coming from -- detail and coring settings, black gamma, the color matrix, the codec, or something wrong with Pancam's individual camera? All my recording up to now has been internal, long-GOP at 50 Mbps -- could that make a difference? And at 30/60 fps instead of 50 fps. I don't see a very noticeable difference in noise between UHD and 1080p footage.

Barry's side-by-side of the old and new firmware convincingly shows an improvement in noise with the 1.25 version. All my tests were done before the upgrade (no time and lousy weather since then!), but still don't look nearly as bad as Pancam's examples.

To Pancam -- I totally get your frustration, and I'm afraid it's quickly reaching a point where you will never be able to trust this camera even if things do improve with a future firmware update or scene files. Sometimes a first impression is so bad that it can't ever be completely overcome. If you're at that point, I hope you can get rid of the DVX200 quickly and find something else that will suit you better. On the other hand, if you still have the ability to "forgive and forget" these rough early days, I think it might just get better soon.

It took almost a year for all the promised features and bug fixes to make it into my PX270, but it has turned out to be a great and trustworthy camera since then.

- Greg

 

[Mike Harvey]

But I have footage from my own DVX200 that has a similar tonal range, including nearly identical black fabrics, and absolutely does NOT have this level of noise in the blacks and shadows. I don't know where it is coming from -- detail and coring settings, black gamma, the color matrix, the codec, or something wrong with Pancam's individual camera? All my recording up to now has been internal, long-GOP at 50 Mbps -- could that make a difference? And at 30/60 fps instead of 50 fps. I don't see a very noticeable difference in noise between UHD and 1080p footage.

If you can't reproduce this level of noise with your camera and the 1.25 update, and if Barry can't, and Pancam takes harddrive and I's advice, and still has a problem, then I'm beginning to think he may simply have a defective unit.

Again, I see the noise, but it's no worse than in my GH4. But if it's far worse than you or Barry or others have, then it may be his individual unit, and he may need to return it for repairs or replacement.

 

[bluesgeek]

I just downloaded Pancam's studio footage and put it in Edius and viewed it on my 20" broadcast monitor in 1080p. From my view I only see a little movement in the jacket. Exactly what Barry is saying in that it is nowhere near the catastrophe that some are making it out to be. Could it be better at 0 db in a studio, yes. Is it horrible? no. Can the AC-160 do a lot better? I would want to see them side by side because what is considered noisy today was whistle clean a few years ago.

There are those looking at this footage in calibrated monitors and seeing only acceptable levels of noise, and there are those looking through consumer monitors or other monitors of varying levels of professional application, and seeing horrendously, distracting swarms of noise. Back in my days as a musician in the studio, the recording and mixdown engineers would have several sets of speakers to monitor playback through, ranging from hi-end monitors to a 3-inch transistor radio box, so that it would be somewhat known what the final mix would sound like to audiences using a variety of listening configurations. It didn't matter if the sound was ideal on a $6000 pair of VonSchweikert VR-4s if it was poor on a $100 pair of Realistic Optimus 5s. Shooting and grading for any monitors other than broadcast is aiming at a moving target. Now it may sound far-fetched, but is it possible that Panasonic engineers are, like Bassman2003 and Barry, not seeing the noise because they are not viewing on lower quality monitors? Just reaching here, but to my eyes on my low-end monitors, the jacket noise pancam's footage is way over the top. If it was mine, I would be doing a secondary on the jacket and crushing it into oblivion.

 

[Bassman2003]

It is still early days with this camera. I am not an owner so I am not defending it but when I look at that footage and then hear "swarms of noise" it makes me think there is something else at play. But I do think that a $4,000 camera should not have any noise at 0 db in a studio lit environment. But this could be setting related. I have a friend you just got this camera and I told him we should set it up one day. So I look forward to twisting the dials and seeing how it compares to my PX270, the low noise champion. (at least for Panasonic video cameras under $20,000).