Sumix 2/3" 1920x1080 CMOS

[Daniel Lipats]

The best way to find out what will be the system requirements for realtime encoding is to just wait for them to release Streampix 3 with cineform and test it out.

The benchmarks done are not on the Sumix camera, but another one so it may be possible that its different.

As soon as Streampix is available Jose and I will run tests and post results. Hopefully I wont have to return this computer after all.

[Juan M. M. Fiebelkorn]

Farhad, thanks for the credit but anyway I wasn't looking for that.I don't think I'm the only one asking for it.

BTW, if you tell your engineers to search the endless posts from around 2004 till today from several people, I guess you will find lots of information and useful resources, from camera features to compression techniques.
I'm mostly interested in getting a "usable" camera.

[Jose A. Garcia]

-10bit RAW 1.8:1 lossless compressed recording with selectable framerate.
-Color chart based automatic callibration with improved white balance.
-Record everything you want as long as you have enough RAM.

If you add future lossless compressed recording to HDD, I prefer Sumix software instead of StreamPix.

Is there a way to change that 1.8:1 ratio to say 3:1 or even 4:1 and still record lossless?

[Farhad Towfiq]

Originally Posted by Jose A. Garcia

Is there a way to change that 1.8:1 ratio to say 3:1 or even 4:1 and still record lossless?

Jose, We understand importance of higher compression ratios. Some form of lossy compression must be acceptable also. When image from 3D world is reflected on the little sensor you already have a tremendous loss of information. Further, silicon in sensor is not perfect and produces more aberrations and loss. logically a little bit more loss should not harm anyone. But in practice at this last stage any little loss may cause big grief. The reason is that traditional lossy compressions do not respect the subtle visual clues in the image used by human eyes to reconstruct 3D, motion, edge, and other important features. Eye must again compress data perhaps 1000:1 or more, if our compression and decompression is not compatible with eye's early vision then human being is annoyed and gets tired of watching.
So, in order not to negatively surprise you later we are not going not make any promise here. Also for you not to be pleasantly surprised we agree that this is a main issue and we would cooperate with our customers in experimenting.
By the way, in our business any surprise must be as much avoided, even good surprise has bad aspects for our customers and our business.

Juan, for you to get a useful camera, consider nothing sacred except continue telling us what you need.

[Gottfried Hofmann]

Originally Posted by Jose A. Garcia

Is there a way to change that 1.8:1 ratio to say 3:1 or even 4:1 and still record lossless?

Consering lossless compression one shouldn't except a ratio of more than 3:1.
Actually, some ration between 2:1 and 2,5:1 is realistic:

http://compression.ru/video/ls-codec/index_en.html

[Juan M. M. Fiebelkorn]

Yes, in fact getting to more than 2:1 requires so much computational power that it doesn't make sense for this type of application.

On the other side I don't see how lossy compression could be a bad thing, despite what Farhad may have said. If that were the case SI's camera and RED ONE would be crap, because they are heavily compressing the Bayer Pattern ranging from 5:1 to something around 20:1.
I understand that for some Industrial/Scientific application that could be a bad thing, but not in our case. We need images to look NICE, not CORRECT :)

The secret is in the details, so you need to know HOW to do it right, you can't just throw your RAW to any compression available and expect good results.I'm saying this because a couple years ago, I coded myself a compression (lossy) for the Bayer Mosaic, which was quite good at the testing stage.

[Jason Rodriguez]

BTW, Juan, in the latest versions of CineForm RAW it's now 3.5:1 compression :)

[Juan M. M. Fiebelkorn]

Ohh, sorry, so I need to correct myself. Starting from 3.5 to anything. Is it Mathematically lossless?

[Ali Husain]

Originally Posted by Jason Rodriguez

BTW, Juan, in the latest versions of CineForm RAW it's now 3.5:1 compression :)

there exists more redundancy in pre-debayered
raw data than with debayered yuv or rgb
data. You'd expect better compression with
raw data. Cineform is great though. :)

[Jason Rodriguez]

Is it Mathematically lossless?

No, not yet, although practically speaking when you get that low with compression ratios, you are on the verge of being "mathematically lossless".

I think for true mathematical lossless compression you would need to be below 3:1, i.e., 2.8, 2.7, etc. It would also be quite a beast processing-wise because you would basically be compressing redundant noise information and reduplicating the noise signal in the image rather than further rendering any "real" image data itself.

So for instance, right now, with the SI-2K, at 2048x1080/10-bit/24P we're hovering around 18.5-20MB/s at 0db gain.

[Noah Yuan-Vogel]

i do recall SI mentioning they were using core2duo mobile processors to record cineform RAW from SI2k. The cutoff was somewhere around 2GHz+ for realtime encoding and preview a year or so ago i think... so im sure a number of the new 45nm Core2duo mobile cpus ought to be up to the task. is it possible those numbers from norpix for quad core encoding are for Cineform RGB rather than RAW? i imagine jason can verify some of this.

as for compression, mathamatically or visually lossless, even a bit lossy would be fine with me as long as it means a real, usable camera solution comes out of it. ram-only recording wont go far, nor will quad core encoding boxes, although i would love to have an excuse to try to find a way to get a new q9300 quad core cpu into a portable computer.

[Daniel Lipats]

Norpix should have something out soon, possibly even this week.

I just hope all this talk about compression and data loss won't lead to compromises that would leave us with a product no better then whats already available cheaper on the shelf.

Please correct me if im wrong but even if its visually lossless it could still cause problems in post with chroma keying or color correction. Also leaving behind artifacts in certain conditions.

The ideal solution for me, and it seems like for others here would be to preserve bayer. Debayer and encode to whatever you want in post. This way there is no data lost and we would have the full benefit of the sensor. But I'm not sure if that's a practical or possible option at all.


However, looking at the Wikipedia page for the RED One has some interesting data:

"Redcode RAW is a variable bitrate wavelet codec which allows raw sensor data at resolutions of up to 4096x2304 to be compressed sufficiently for practical on-camera recording. Two variants are offered, one with a maximum data rate of 28MB/s (224 megabits), and one with a maximum data rate of 36 MB/s (288 megabits). Compared with the uncompressed data captured by the sensor, these bitrates represent compression ratios of about 12:1 and 9:1, respectively.

Because Redcode is a wavelet codec, similar to CineForm RAW and JPEG2000, the blocking artifacts associated with other digital video compression algorithms are absent. However, sample images detailing Redcode specific artifacts have been posted on the manufacturer's forum. References to such problems existed until very recently when the message threads containing this information were removed by the Reduser.net administrators.

Redcode is a mathematically lossy codec, meaning that decompression does not fully restore the original image data captured by the camera. Red does claim the codec is "visually lossless", meaning that the information loss is not visible to the naked eye when images are viewed, however this is a subjective statement."

So, maybe compression really is a necessary evil.

[Gottfried Hofmann]

Originally Posted by Daniel Lipats

So, maybe compression really is a necessary evil.

No, it's not. All you need is bandwidth, storage and a powerful CPU ;)

The most problems with compression come up in post, not when you look at your recording.

In my opinion there should always be the option to get the uncompressed raw data with maximum bit depth. Mathematically lossless compression of course is a good thing.

All the stuff like lossy compression or RBG conversion with LUT can be added as a feature.

[Farhad Towfiq]

Time is on the side of less compression and lossless compression as computer technology advances. Lossy compression in combination with image enhancement must be done only in post for distribution. Placing sensors for capturing signature of vibrations can be used to remove undesired mechanical movement in video only if you have raw data.
By the way,the same way equipment and fan noise can be recorded and used to remove the parasitic noise from audio. So there must be no fear of fast HD noise, fan, etc.
If Red one is doing 9:1 compression then SMX-12A2C will start to beat it in overall quality.

[Jason Rodriguez]

Please correct me if im wrong but even if its visually lossless it could still cause problems in post with chroma keying or color correction. Also leaving behind artifacts in certain conditions.

Sean Hellfrisch and Isaiah Saxon just completed their Bjork 3D music video with all green-screened elements. We also have a number of green-screen elements on our website that you can play with.

While the elements on the website and what Sean and Isaiah produced with were at 5:1 compression (we didn't have the 3.5:1 update from CineForm yet), we think the results are still really good.

Also for comparison, the HDCAM-SR 444 RGB mode at 440Mb/s is more compression than 3.5:1 CineForm . . . and HDCAM-SR 444 RGB mode is very high quality. So I think for the price, you will not find anything delivering the compression quality of CineForm in any other off-the-shelf solution (that can also be edited in real-time on the same box you're recording on . . . using J2K at 3.5:1 compression is not a real-time editing solution).