Rumored nikon d90 buffer upgrade

[beckspace]

good, now we're getting somewhere. now you just have to compare that to the same shot using the d90 dmovie mode, or just check out that other thread already since people have already been doing it for you.

EDIT: http://www.dvxuser.com/V6/showpost.php?p=1458475&postcount=5

/matt

I don´t do any flame war here, but I can assure you, using my socalled clinic eye, that all artifacts, besides the steps, are made by JPEG low bitrate compression. JPEG images are what DSLRs do since they came to existence, those chips are created for JPEG compression, but the D90 keeps it low. You blamed the image processing "noise reduction, debayering, scaling, whatever", no, it all dont harm the image at anything. The image processing (at 16 bits before it is compressed to 8 bits JPEG) is what Nikon do the best, the problem IS the compression

 

[mattsand]

yeah, i didn't want to start a flame war either but now i'm starting to feel like it. you're an idiot, to use the words of another arrogant sob with a famous clinic eye. i won't venture further into this but take a look at that god damn link already, or test it yourself. i've given up but maybe you can save yourself some embarrassment in the future...

/matt

 

[USLatin]

Ok, guys, please knock it off before the mods have to come and yell at people. You are both trying to figure out the same thing. Let's carry on the conversation in a bit more respectful manner as the rest of us don't want to have to read insults. Thank you.

Not to hack the thread, but since this is along the lines... does anyone know if Nikon's firmware was ever hacked at all? For any camera? If someone could figure out the max sustained writing rate to the cards and tested some stuff to see if the processor could handle it we could even see other real fancy footage coming out into the cards... sounds complicated, but has anyone out on the web try to modify Nikon's firmware in the past? I keep coming up with bogus threads that lead nowhere.

 

[beckspace]

I believe the D300 was hacked, giving some bracketing functions and more fps

 

[USLatin]

I saw something about that too, but it seems it was just a few values that were changed, nothing fancy. Would you be able to link me though? Maybe what you found was more elaborate than what I was able to find.

All it takes is one programming geek and one photography hardware geek to sit down for a cup of coffee/red-bull and figure out what the possibilities are... well they would have to test a bit too so it might be a couple of beverages.

The processor of the D90 is the same as the one in the D300, so it should be able to handle some serious processing... I believe the D300 shoots 8sps (stills per second).



Ok, back on topic: I'd love to be able to check out your comparison with a day-time picture.

 

[beckspace]

yeah, i didn't want to start a flame war either but now i'm starting to feel like it. you're an idiot, to use the words of another arrogant sob with a famous clinic eye. i won't venture further into this but take a look at that god damn link already, or test it yourself. i've given up but maybe you can save yourself some embarrassment in the future...

/matt

Hey Matt, you gave up, but I didn't.

I read his thread, but one thing escaped him: one way to create a JPEG to fit in a narrow bandwidth is to blur it. Its called the JPEG Blur

To emulate a D90 D-Movie frame, take a clean 1280x720, Save for the Web, Put JPEG Blur at 0.6 or 0.8 (it changes due to the complexity of the image), and then change quality to fit into a 104K image, look at this:

The Blur is from the D90 compression trying to fit the image into the 20mbit/s at 24fps avoiding terrible JPEG artifacts

do you need any other evidence?

 

[USLatin]

It sounds like you are on to something, but please don't make it sound like an episode of CSI Miami

SO I take it that you were able to fit the full 1280 x 720 image into the 104k and these are just 100% crops of that, right?

Could it be then that the D90 just uses a stronger blur setting to make sure that all images fit regardless of how much detail they have? I would suppose having to evaluate them before applying the blur might take up a lot more processing power hence they found an arbitrary level that works 100% of the time... I can imagine Photoshop's blur being better but the 0.9 is certainly nowhere near as blurred as the D90's...

is there a way to find the actual size of each frame to see if they end up being even smaller than 104k? that would support my theory on your theory...

 

[beckspace]

It sounds like you are on to something, but please don't make it sound like an episode of CSI Miami

sorry

SO I take it that you were able to fit the full 1280 x 720 image into the 104k and these are just 100% crops of that, right?

yes

Could it be then that the D90 just uses a stronger blur setting to make sure that all images fit regardless of how much detail they have?

without the blur the compression start to make the damned blocks, the blur gets stronger with complexity, which I saw in various landscape shots. If the bitrate was just a little bit higher, lets say 160K per image, we would see very different images

is there a way to find the actual size of each frame to see if they end up being even smaller than 104k? that would support my theory on your theory...

I did not take account of the noise from the sensor, and yes, the D90 frame seems to be even smaller than 104K, 104K is optimistic

 

[USLatin]

without the blur the compression start to make the damned blocks, the blur gets stronger with complexity, which I saw in various landscape shots.

So I guess that it looks like the camera actually applies different amount of blur to each frame and it actually analyzes each frame in some sort of way, possibly looking at a histogram to see if there is a lot of high frequency detail or something of the sort..?

 

[beckspace]

So I guess that it looks like the camera actually applies different amount of blur to each frame and it actually analyzes each frame in some sort of way, possibly looking at a histogram to see if there is a lot of high frequency detail or something of the sort..?

I don´t know, I think it first compress, then if it gets too big, it apply the blur to fit, its more than one process in real time, I guess. maybe thats why it doesnt lock the exposition out of the box, I saw that when you press AE Lock, it also locks the Blur

If thats true, a new software "buffer" could make a hell of difference

 

[bearing]

There is no such thing as a "JPEG blur". Photoshop is just blurring the picture with a gaussian blur or something, before applying JPEG compression.

As I wrote in the post - the difference between the 100% quality picture and the 54% picture is negligible. Yes, you can see some blockiness and some some bluring - but compared to the d-movie frame it's nothing. I can upload the one with 100% quality (1MB) if you want to compare.

 

[bearing]

I think the reason to the blurriness is the debayering stage when the sensor has been read through skipping.

Normal debayering is done similar to this:

Debayering when sensor has been read through skipping is using sensor elements far from the destination which will look like blurring.

 

[Lee Wilson]

Knight to Bishop four.

Did I win ?

 

[USLatin]

Wow.. yea that would make a lot of sense too bearing, but not sure if it would create just a soft blur like a Gaussian blur would... so maybe it is a combination of both... a weird debayering like picture #2 on bearing's post which might create some weird artifacting stuff, plus the deliberate blur in-camera... it does look like there is a hell of a lot of stuff going, certainly something beyond a blur when you look at the water...

I don't know why I am reading so much about this cause I am not planning on using a D90 for video, opnly day-dreaming about having one for stills... but let's hope they make it better cause it would be even more of a sweet sweet piece of equipment if the video were more acceptable. Certainly a sweet toy to own.

PM me when you get to the bottom of this puzzle... I imagine it might take a few more pages at least anyway.

 

[beckspace]

when I meant "JPEG Blur" is about the most used technique to shrink down the file size of a picture, not a built-in Blur
If your Debayering theory was right, we would see it in every image that this camera produces, and that´s not the case.



But let´s take a real life example here:

I took this shot with AE Lock pointing to the sky, the simplest image available that anyone here can verify with their D90s,
you´ll see it doesnt apply the Blur, you can even see a bird flying with no artifact around it

But then, in this same shot, in the same AE Lock, from the sky to the mountain, from a simple to a more complex image, look what happens:

The AE Lock also locks the Blur in the compression, so it hasnt the Blur, but instead the JPEG blocks appears because there is no bandwidth for this image


You can test it, try it. I just don´t understand why my opinion is being so mined,
I´m just applying some Occam´s Razor here, and the simplest theory for the D90 artifacting is the Mjpeg Compression at 20mbit/s.


We should be more open minded around here and help each other instead of cutting throats.
We are pongs, not kings, we are at the bottom of the food chain of moviemaking,
pongs should help each other, because no one else does

 

[bearing]

>If your Debayering theory was right, we would see it in every image that this camera produces, and that´s not the case.

No, we should not. Why should we?
When taking stills all 12M pixels are of course used, without skipping, like on the first picture.

It's unlikely the sensor can read 12 mega pixels 24 times per second. One can assume that the maximum is 4.5 fps since that is the maximum fps of still images. If that is the case the readout frequency is about 12*4.5 MHz which is 54 MHz. With a 54 MHz readout frequency the maximum image size at 24 fps is 54 / 24 MP which is 2.25 MP. 1920x1080 is about 2.1 MP...

 

[beckspace]

If think besides Nikon no one knows if the D-Movie ignores columns and rows of pixels to generate a 720P, but even if it does, it cannot be responsible for those artifacts. Debayering does not create blocks or such a blur (even if it only transform RAW into RGB without downsampling), and as I said before, the artifacts change its behavior on each exposition, that by logic, it can only be attributed to the compression

And my guess is D90 scans the full sensor all the time, as in Photo, as in LiveView and as in Video. But that is guessing

 

[Alpha Beta Omega]

The D90's JPEG compression varies greatly. From 2/Mbps to +30/Mbps. You can't expect great things with 2/Mbps but when it jumps up to 15/Mbps and above, it's usable.
The thing that dictates the bitrate is the scenes complexity but mostly it's the light coming in. More light-higher bitrate, less light-lower bitrate. So for the highest bitrate-good exposure with good light (lots) and lots of detail.

JPEG compression as in the D90 throws away everything thats to poorly lit and black is the worst. It makes the bitrate fall as a stone.

 

[bearing]

when I meant "JPEG Blur" is about the most used technique to shrink down the file size of a picture, not a built-in Blur
If your Debayering theory was right, we would see it in every image that this camera produces, and that´s not the case.

Ok, you meant every moving image it produces. You may be right, but I don't think so. For starters - when I took my pictures I locked exposure with a white balance cap in front of the lens. If your theory about locking exposure is also locking the blur is right - my pictures wouldn't be blurry, but they clearly are.

you´ll see it doesnt apply the Blur, you can even see a bird flying with no artifact around it

The bird looks pretty blurry IMHO.

But then, in this same shot, in the same AE Lock, from the sky to the mountain, from a simple to a more complex image, look what happens:

Quantization in JPEGs has increased. I honestly can't see any increase in blur.

You can test it, try it. I just don´t understand why my opinion is being so mined,
I´m just applying some Occam´s Razor here, and the simplest theory for the D90 artifacting is the Mjpeg Compression at 20mbit/s.

We should be more open minded around here and help each other instead of cutting throats.

I'm sorry if I sound like I'm approaching you with a knife. I think the reason is that you are stating things as facts.

To change my mind you have to show me evidence and arguments I can believe in. I did my testing in a scientific way. I locked exposure to ISO200 1/50 every time. I shot from a tripod. I did the same test many times to have like a statistic base. I watched the quantizing tables change as I slowly removed the lens cap.

 

[beckspace]

I can´t think of any other evidence to show, my background is photography, not video

I showed that 100K (20mbps) noiseless images are not enough to film 720P images with too much information

that a JPEG Blur creates the same artifact as D-movie filming complex images (to avoid macroblocking, thats why Bearing forest shot in DMovie is so unacceptable blurry)

that AE Lock also locks the compression settings (JPEG Blur) depending on the image complexity (and maybe light as editman said)

think in how DSLRs are assembled, there is a reason for D90 opted for Mjpeg, JPEG is what DSLRs do, so Nikon kept it simple. There is a Sensor, an Image Processing (RAW) and then the Compression (8bits conversion), nothing more

One blamed the Sensor (bearing) suspecting that it skips pixels (?), other blamed the Image Processing whatever (matt) and I blamed the Compression bandwidth with its JPEG Blur flattenning the black, moisturing the noise, eliminating fine details to avoid the macroblocks in such a low bitrate for 720P Mjpeg

just got tired in doing this, so let me rest my case.

If there is a buffer update, we'll see if I got all this so wrong

And Bearing, you were respectful, others weren't so much