Here Untitled Document (http://evengod.eu/shoot/pics2.html) you can see a couple op pngs in full res that i took from the window of my studio. One is recorded on the SxS in 1080p HQ and the other captured with a Blackmagic in uncompressed 4:2:2 10 bit. I made the test very fast after receiving the camera back from the service for a firmware upgrade and i didnt realize that it was some picture profile on. I will make some more detailed tests and post it. So far I can see that if I zoom in on the pics, the one from the SxS shows some mosquito noise that is not present in the other from the SDI. I didnt have time to look much more but you can tell your opinions.

Um, I'll take your word for it. For my money, Long GOP Mpeg is pretty impressive, comparing it to SDI. I really don't see much difference!

Thanks for this!

It's hard to tell the way it looks in my browser as a still vs. playing back in motion on a calibrated monitor. Of course it's a no-brainer that the SDI should have less apparent compression artifacts compared to 35Mbit XDCAM HD. That said this is one of the best codecs I've seen yet in that infinite trade-off between efficiency, storage requirements and image quality.

-Noah

Mosquito noise in particular shows up more with a moving image as the noise varies from frame to frame.

The difference between the compressed and uncompressed material is small. I have done similar tests and visually the difference is very small. However there are artifacts introduced by the compression and no matter how small every de-code, re-encode process that your material may encounter from the edit to final viewing will add artifacts on top of artifacts. It is this concatenation that can lead to bigger problems further down the chain.

The MPEG produced by the EX cameras is very, very good and if handled carefully will produce stunning results. However if you can utilize the uncompressed output using something like a Flash XDR you will have material that should be more robust in post production.

Mosquito noise in particular shows up more with a moving image as the noise varies from frame to frame...

The "busy pixels" are only busy with moving images - comparing stills doesn't make sense to me.

if you like when i make new tests i can post the footage instead of stills

if you like when i make new tests i can post the footage instead of stills

Please do; would be of great interest - especially to those who're planning to buy FlashXDR.

Which codec did you record to?

[edit] Pays to read the first post, 10 bit!

...I can see that if I zoom in on the pics, the one from the SxS shows some mosquito noise that is not present in the other from the SDI.It's pretty a much a given that SDI is superior, but think about it: the average person viewing your material is *not* going to zoom in on the pics. I think Alister says it best by pointing out that "the MPEG produced by the EX cameras is very, very good and if handled carefully will produce stunning results." Enough said...

May be there is no point to discuss if SDI is better or if the difference it would be seen, etc. Each tool is good or not for the intended work. I´m sure that for chroma keying or for extreme post treatment (call color grading or whatever) SDI will make a huge difference. To film my kid playing at the playground...well, of course i wouldnt take the extra work and cost that comes from shooting uncompressed. Anyway is good to know how much each system can give and that the whole point. I´ll continue the research and post the results.
Personally i work a lot making compositing and complex post-processing and sometimes i would give my entire kingdom for a better footage without noise, better color resolution, deeper chroma sampling and so on.

Bob, the codec was Blackmagic 4:2:2 10 bits RGB uncompressed

I'll continue the research and post the results.
Yes, that is the main thing -- and please do so by all means. Many thanks for all you've done so far.

The SDI improvement can be seen on the surface of the water. On the other pic it's not mosquito noise, it's mpeg block noise.

Also, notice the dynamic range. I don't know if SDI bypasses the gama curves, but if you look at the side of the car in the SDI photo, it is much brighter and a lot more detail available around the wheel than on the MPEG picture. You see the same thing around the rusty bolts at the base of the column - you can see into the shadows there better in the SDI picture.

The difference between the compressed and uncompressed material is small. I have done similar tests and visually the difference is very small. However there are artifacts introduced by the compression and no matter how small every de-code, re-encode process that your material may encounter from the edit to final viewing will add artifacts on top of artifacts. It is this concatenation that can lead to bigger problems further down the chain.

Unless you convert all your footage to uncompressed or a lossless codec like ProRes. Then you can color correct and do effects without degradation. As good as the EX1 MPEG may be, MPEG is for acquisition only, like HDV. If you have anything more than just straight cuts you have to convert it to uncompressed or ProRes if you want no degradation.

Also, notice the dynamic range. I don't know if SDI bypasses the gama curves, but if you look at the side of the car in the SDI photo, it is much brighter and a lot more detail available around the wheel than on the MPEG picture. You see the same thing around the rusty bolts at the base of the column - you can see into the shadows there better in the SDI picture.

That would be the 10 bit quantisation of HD-SDI vs the 8 bit of MPEG2. 1024 levels vs 256. It doesn't bypass the gamma circuits, the HD-SDI signal can just carry more of the original information. I think the CMOS chip outputs 14 bit, which is 16384 levels, which gets averaged down to 1024 or 256.

I didn't have enough time to make more footage (i will i promise) but may be this is an approximation to the definitive answer. make this:

download both images. One comes from the SxS in 8 bits and the other from the SDI in 10 bits and both images are PNGs 16 bits so there is no doubt about how many colors the files can handle.

Make a levels adjustment layer in every picture. Looking at the histogram you already see the differences, the SxS have a lot of white bands (areas that has no information).

Now let push things really far: make a severe levels adjustment like 80 1 160.
What do you see? yep, the SxS image have a lot of banding (even if this image is not the best to judge banding since it doesnt have a big area with a soft chroma or luma transition but anyway you can see the banding) and noise (imagine that noise in the moving image, i mean, the noise moving from frame to frame, Ughhh!!)

Now turn off the levels adjustment layer and make a new adjustment layer with Hue/Saturation, turn the saturation up to, lets say, 70. What do you see? the SxS image have a lot of color noise and banding too.

I know, you will say, "I seldom have to make so severe color correction" but take your time and play a little with the pics and you will see that in many cases an enhancement in contrast is enough to show a big difference between the images. In the other hand many times you want to attain a look that needs severe color correction (and not this useless levels and saturation test).

if you are unsure because of the difference in brightness go to match colors on the SxS file and match to the SDI and then make the tests (don´t make the opposite since the SDI have more info and you will downgrade the image)

if you have enough patience try this: duplicate the image layer, on the top layer put a high-pass filter with a radius of 3 pixels (you can use another value if you like), set the blending mode of this layer to overlay. I used this technique many times to sharpen an image without introducing too much noise, then you can play with the opacity of the layer to set the amount of sharpness that you like or even mask out the areas where you don´t like to have the sharpness. as you can see on the SxS image the noise is sharpened as well (in fact in both images the noise is sharpened but the SDI image have less noise). Again, if you don´t plan to make strong corrections to your footage the visual difference is not really too much to make it worth to work in uncompressed, sdi, etc, otherwise...

Also consider that many times to make some types of looks you will be doing in several passes, even with several different softwares and each pass will show more and more the limitations of the compressed video.

Now if you have access to a color grading package like Apple Color or Autodesk Lustre, or whatever, open both images and try to qualify for a secondary. As you can guess the SxS is much more difficult to qualify. So many times you will find yourself making masks to be able to make secondary color corrections. In this particular image this issue is not big deal but, what happens when your camera is moving (with changes on perspectives) and character passing in front of the object or area you want to make that secondary CC? it could be a nightmare.

That´s why a device like the nanoflash is more than welcome for me. The only thing I wonder is: since 4:2:2 footage takes much more space than 4:2:0, would it be 100mb/s 4:2:2 much less compressed than 35mb/s in 4:2:0? no answer so far, i have to take a closer look to the numbers.

That´s why a device like the nanoflash is more than welcome for me. The only thing I wonder is: since 4:2:2 footage takes much more space than 4:2:0, would it be 100mb/s 4:2:2 much less compressed than 35mb/s in 4:2:0? no answer so far, i have to take a closer look to the numbers.

I've calculated that a 4:2:2 image holds 20% more raw information than a 4:2:0 image of the same. You can't really say then though from that, that you need a 20% higher compression rate to maintain the same fidelity. Compression is a very complex thing as you know I'm sure. It would change from image to image, but likely less than 20% extra on average I'd think.
The colour fidelity of 4:2:2 is a large leap from 4:2:0 though, with one out of two pixels the correct colour, rather than a block of four averaged to "almost" correct.
It's a pity Sony are so bent on protecting their high end market, as this camera could easily have recorded 4:2:2 to SxS natively at no greater cost.

It's a pity Sony are so bent on protecting their high end market, as this camera could easily have recorded 4:2:2 to SxS natively at no greater cost.

I agree, so for a little extra hardware cost, I will be able to 4:2:2 record from my EX3 onto my upcoming nanoflash. Looking forward to that in the next couple of months or so.

After the XDR & NF are out, Sony will need to rethink their 4:2:2 pricing, just as you have suggested. An EX5??? upgrade maybe???

Cheers
David Issko