Chinese 2K RAW S35 Cinema Camera, 8000 dollars

Not the Epic-M. The M was advertised as being released as-is, with more features to come in the future. Potential buyers were advised as to the state of the "M" project before buying.
 
TimurCivan said:
Now riddle me this. Had the tsunami never happened, and epics were produced this spring on schedule, Jannard stated that they would have fully functional firmware when they shipped epic x. Meaning a finalized firmware should have been available in april.
Click to expand...

Personally, I think speculating about the past is even less useful than speculating about the future since we are all aware that any given certain turn of events did not occur in any way other than that which is currently presenting us with "reality". But, sure, why not?

Well, I don't believe Jim said it would be "finalized" or even "fully functional" - just that they would be shipping a camera that worked out of the box without the need for signficant firmware revisions, as occurred with the RED ONE by design.

That being said, why couldn't they have just finalized the firmware in April? Firstly, I have no true knowledge of that decision. And I don't think anyone else here does unless Jim or Jarred decides to chime in.
  1. It's very possible that they had to delay the firmware development because they were no longer sure what hardware they would be using.
  2. It's also possible that they had to delay the firmware development because they knew what hardware they would be using, but couldn't get enough of it because of the tsunami to effectively develop.
  3. It's also possible that they simply decided, since the hardware was delayed, to delay firmware development until closer to the hardware date and put their developers onto other tasks, like the canon mount, REDMOTE, etc.
  4. It's also possible that they couldn't have shipped anything close to "final" or "finished" or "working" or whatever you want to call it back in May or June.
  5. It's also possible that they could have shipped a perfectly working camera in May or June, though without some of the extras we are now seeing in Epic-M.
So take your pick. Either way, reality is what it is. We are getting Epic-Ms now, Epic-Xs very soon. And they are apparently "stable" or "finished" or "working" for Vilmos Zsigmond to announce he's shooting his next project with it, for John Schwartzman to shoot The Amazing Spiderman with it, Geoff Boyle seems to have something in the works, The Girl With The Dragon Tattoo, Contagion (I believe is being partly shot on Epics), Ridley Scott's Prometheus, Don McAlpine on Mental, as well as Peter Jackson's The Hobbit. Those are not what I would consider "low-risk" productions where its fine if a camera goes down for a few days.

John Schwartzman was kind of enough to make several posts over on REDUSER about using the Epic-Ms on set, and described the experience. He was very happy, and apparently dealt with very very few technical difficulties.

TimurCivan said:
And.... for 58k, and being hand bult it should be better, not worse.
Click to expand...

Well, that depends. A machined, hand built high-performance engine from Ferrari may have some kinks to work out if it is intended as a sort of prototype. Which EPIC-M is. And it says so, right on the product page...nothing hidden about that. A machined, hand built production engine, though, should be spectacular.

And, ultimately, I think the idea of a "finished" camera is changing. Look at the Alexa. Look at the F3. Both added significant updates after launch...months after launch. Is this better...maybe not.

Regardless, this is getting off topic - we may soon have another camera in the game!
 
TimurCivan said:
That was the whole point of the epic. No more beta. A finished camera.
Click to expand...
I guess "final" is a subjective term. Certainly, Epics -- including M -- have more features on release than did Red One, but I was under the impression that not all features will necessarily be completely finished on release of the Epics or Scarlets. I believe that I read Scarlet won't support both forms of HDRx (precombined and two separate streams) out of the gate, but I may be misremembering.
 
Wait, is this the RED EPIC thread or the Chinese 2K thread? Cause I just waded through pages of epic discussions:zombie_smiley:
 
So let me get this straight. This company is developing a S35mm CMOS sensor camera, a S16mm CCD sensor camera and a S8mm CMOS sensor camera, all with interchangeable lenses and variations on RAW recording to internal SSD drives. AND the larger camera is projected to have a $8000.00 price tag. That is crazy. I love it... "I do believe in fairies, I do, I do... clap your hands Wendy, Clap your hands"!!!!
 
Dan, I don't know if you are allowed to answer this one or not, but...

Is this camera shooting on a 2K chip and then giving 2K Raw files?

Or is this camera shooting on a larger chip, downsampling in the camera and then somehow exporting that as raw?

If the former, I'm curious to know what sort of hit they are taking in resolution because of the debayer. I'm aware that given the right software, it very well may improve to be upwards of 80% effective.
 
:shocked:
Dan Hudgins said:
It would be good to do some reading on anti aliasing issues with computer graphics.Video cameras have used peaking circuits to get over rated resolution value, but since there was noise in the sweep and sync circuits and the electron beams did not focus to a fine point the aliasing was not as noticable perhaps as in digital images were the raster stands still rather than moving around and so gainiing some "sub-pixel" resolution. Also in analog systems you can get 1 pixel width, but you can also get 1.1, 1.2, 1.3, 1.4, 1.5, 1.75 and such all the way up to 2 pixels width so its not a gross jump at the HF end of the detail range like in digital capture and display, that also can reduce aliasing.Square pixels are 1.414 time as long on the diagonal, so if you take the maximum resolution of a digital image as 0.707 on the height and width of the pixel count, then the diagonal must be 0.707 times 0.707 or about 0.5 measured in H or V units.To mimimize aliasing the 80% MTF resolution can for the most part be no higher than 25% of the horazontal pixel count in ANY digital image with square pixels on all angles that a line edge can take in the image, roughly. So a 4K sensor gives about 1K visual resolution. If you push the sharpness higher then you may see some runners or other aliasing artifacts in slow moving images, such as a test chart rotating on a turntable mounted on a wall in front of the camera.THe Bayer filter brings in its own problems as you need a OLPF to span the 75% gaps in red and blue and the 50% gaps in green.Anyone building a camera faces the same issues with aliasing, chroma moire, and how much anti-OLPF sharpen you can apply before the aliasing becomes ugly or an issue.My software DANCINEC.EXE (tm) has adjustments for filtering and adjusting the anti-OLPF sharpen. Because the OLPF are about 2mm thick they introduce some "plate effect" due to the fact that many film lenses are not corrected for that thickness of glass behind the lens. So that 2mm thickness is going to blur maybe a bit more on the S8 camera than the S16 camera and even less on the S35 camera in relation to the image height.These Engineering issues are a point of discussion because they relate to de-Bayer software development and in some ways may be greater than the sensor selection in their impact on the image quality.

35mm movie film negatives have a upper "resolution" limit at small MTF values of 4K to 6K or so, but each generation prints the resolution lower since by contact or optical printing the result cannot be as sharp as the original without un-sharp masking through bi-pack. So after four generations, camera negative, master positive, duplicate negative, then release print the "resolution" is down to about 1.5K to 1K, and then the projector lens acts like a low pass filter and can be out of focus so the lines per image height at 80% MTF of a 35mm movie on a theatre screen ends up about between 960 and 512 lines per image height of visual resolution from someone sitting 2/3 back from the screen.

Its not right to take high extinction values like 0.8 of horizontal pixels near 0% MTF as "useful" resolution that will make it from the Digital Cinema Camera through the 35mm film recorder, to the internegative stock, then printed to the release print stock on high speed contact printers, then projected out of focus with an older projection lens in theatre, as each step will push those low MTF at high spatial frequencies so low that their contrast makes them invisible or at least not significant. So called 2K filmouts are done at 1828 wide sometimes as there is a loss of pixels from the analog sound track on the side of the image. 1828*0.7 = 1280.

What matters over all that is more important for a useful is tonal range and texture in the color pallet of the gambit captured, highlight detail, and other image aspects that are not "resolution" dependent so much perhaps.

If you apply too much anti-OLPF sharpen to get 80% "resolution" from the horizontal pixel width in the de-Bayer, then you end up with the over sharpened "video" look rather than the soft "film look" if you have the bend point too close to the high spatial frequency first null, that is does the curve droop from 100% MTF at 1K or does it go up to greater than 100% MTF at 1.5K then drop back down to 1.8K like it would in an over sharpened video camera? Such over peaking can give high values for a spec sheet, but its what is on the movie screen that matters to a filmmaker, I would hope.

Arri Alexa (tm) and RED Scarlet use 2880x1620 on the Bayer pattern and then de-Bayer and downsample to 1920x1080, that can improve the images since some de-Bayer artifacts null with the 1.5x reduction, if you do that reduction before the de-Bayer the gain from the null is not gained. If you de-Bayer in the camera at the oversample rate then reduce and save the RGB image at the smaller size, the recording bandwidth is higher,

2880x1620*1.5*24 = 167,961,600 bytes per second.

1920x1080*1.5*3*24 = 223,948,800 bytes per second.

So the Bayer at the 1.5x oversample requires a lower recording bandwidth than does the RGB resize and recording in the camera.

The KineRAW-S8 records at a bit over 2.5K and my DIY film recorder at about 1600 (using a 1600x1200 LCD with extra bright back lighting), so there is a good oversample ratio for my needs there. I going direct to the color print stock without a negative, so the losses from the negative and its printing are not involved so the end resolution is about like using a 1828 crt film recorder perhaps from skipping the negative and printing and recording direct from the LCD panel, the trick is to take the monitor apart and use a large fresnel lens behind the LCD pannel to focus a bright heat filtered light source direct into the 35mm animation camera's lens to get faster exposure on the about ASA 0.5 color print stock.) The current revision of DANCINEL.EXE (tm) is on my web site if anyone wants to bild their own film recorder, it has special gamma compensation to allow for a negative image on the monitor to give the right positive image on the print stock, I hope to make some improvements as time goes on, but its been tested and the results look usable so far with the right adjustments and three pass printing (to compensate for the 8bit limits of the video card and the LCD monitor circuits).

Details of the S35 camera are not finalized as far as I know, so you should wait for published specs on the Kinefinity.com web site.
Click to expand...

I really think that you should consider being more thorough with your responses, Dan. :shocked:
 
Dan Hudgins said:
It would be good to do some reading on anti aliasing issues with computer graphics.Video cameras have used peaking circuits to get over rated resolution value, but since there was noise in the sweep and sync circuits and the electron beams did not focus to a fine point the aliasing was not as noticable perhaps as in digital images were the raster stands still rather than moving around and so gainiing some "sub-pixel" resolution. Also in analog systems you can get 1 pixel width, but you can also get 1.1, 1.2, 1.3, 1.4, 1.5, 1.75 and such all the way up to 2 pixels width so its not a gross jump at the HF end of the detail range like in digital capture and display, that also can reduce aliasing.Square pixels are 1.414 time as long on the diagonal, so if you take the maximum resolution of a digital image as 0.707 on the height and width of the pixel count, then the diagonal must be 0.707 times 0.707 or about 0.5 measured in H or V units.To mimimize aliasing the 80% MTF resolution can for the most part be no higher than 25% of the horazontal pixel count in ANY digital image with square pixels on all angles that a line edge can take in the image, roughly. So a 4K sensor gives about 1K visual resolution. If you push the sharpness higher then you may see some runners or other aliasing artifacts in slow moving images, such as a test chart rotating on a turntable mounted on a wall in front of the camera.THe Bayer filter brings in its own problems as you need a OLPF to span the 75% gaps in red and blue and the 50% gaps in green.Anyone building a camera faces the same issues with aliasing, chroma moire, and how much anti-OLPF sharpen you can apply before the aliasing becomes ugly or an issue.My software DANCINEC.EXE (tm) has adjustments for filtering and adjusting the anti-OLPF sharpen. Because the OLPF are about 2mm thick they introduce some "plate effect" due to the fact that many film lenses are not corrected for that thickness of glass behind the lens. So that 2mm thickness is going to blur maybe a bit more on the S8 camera than the S16 camera and even less on the S35 camera in relation to the image height.These Engineering issues are a point of discussion because they relate to de-Bayer software development and in some ways may be greater than the sensor selection in their impact on the image quality.

35mm movie film negatives have a upper "resolution" limit at small MTF values of 4K to 6K or so, but each generation prints the resolution lower since by contact or optical printing the result cannot be as sharp as the original without un-sharp masking through bi-pack. So after four generations, camera negative, master positive, duplicate negative, then release print the "resolution" is down to about 1.5K to 1K, and then the projector lens acts like a low pass filter and can be out of focus so the lines per image height at 80% MTF of a 35mm movie on a theatre screen ends up about between 960 and 512 lines per image height of visual resolution from someone sitting 2/3 back from the screen.

Its not right to take high extinction values like 0.8 of horizontal pixels near 0% MTF as "useful" resolution that will make it from the Digital Cinema Camera through the 35mm film recorder, to the internegative stock, then printed to the release print stock on high speed contact printers, then projected out of focus with an older projection lens in theatre, as each step will push those low MTF at high spatial frequencies so low that their contrast makes them invisible or at least not significant. So called 2K filmouts are done at 1828 wide sometimes as there is a loss of pixels from the analog sound track on the side of the image. 1828*0.7 = 1280.

What matters over all that is more important for a useful is tonal range and texture in the color pallet of the gambit captured, highlight detail, and other image aspects that are not "resolution" dependent so much perhaps.

If you apply too much anti-OLPF sharpen to get 80% "resolution" from the horizontal pixel width in the de-Bayer, then you end up with the over sharpened "video" look rather than the soft "film look" if you have the bend point too close to the high spatial frequency first null, that is does the curve droop from 100% MTF at 1K or does it go up to greater than 100% MTF at 1.5K then drop back down to 1.8K like it would in an over sharpened video camera? Such over peaking can give high values for a spec sheet, but its what is on the movie screen that matters to a filmmaker, I would hope.

Arri Alexa (tm) and RED Scarlet use 2880x1620 on the Bayer pattern and then de-Bayer and downsample to 1920x1080, that can improve the images since some de-Bayer artifacts null with the 1.5x reduction, if you do that reduction before the de-Bayer the gain from the null is not gained. If you de-Bayer in the camera at the oversample rate then reduce and save the RGB image at the smaller size, the recording bandwidth is higher,

2880x1620*1.5*24 = 167,961,600 bytes per second.

1920x1080*1.5*3*24 = 223,948,800 bytes per second.

So the Bayer at the 1.5x oversample requires a lower recording bandwidth than does the RGB resize and recording in the camera.

The KineRAW-S8 records at a bit over 2.5K and my DIY film recorder at about 1600 (using a 1600x1200 LCD with extra bright back lighting), so there is a good oversample ratio for my needs there. I going direct to the color print stock without a negative, so the losses from the negative and its printing are not involved so the end resolution is about like using a 1828 crt film recorder perhaps from skipping the negative and printing and recording direct from the LCD panel, the trick is to take the monitor apart and use a large fresnel lens behind the LCD pannel to focus a bright heat filtered light source direct into the 35mm animation camera's lens to get faster exposure on the about ASA 0.5 color print stock.) The current revision of DANCINEL.EXE (tm) is on my web site if anyone wants to bild their own film recorder, it has special gamma compensation to allow for a negative image on the monitor to give the right positive image on the print stock, I hope to make some improvements as time goes on, but its been tested and the results look usable so far with the right adjustments and three pass printing (to compensate for the 8bit limits of the video card and the LCD monitor circuits).

Details of the S35 camera are not finalized as far as I know, so you should wait for published specs on the Kinefinity.com web site.
Click to expand...

mind=blown

I just learned so much.
 
Dan you have been super helpful, I have one question... is there a projected price point for the s16 cam that you are aware of? this is the camera that catches my eye, I have worked with the A-Cam, great camera, but expensive and unreliable... if this camera is less than the s35, and shoots .dng footage that looks like the a-cam and is full measurable 1920x1080... Im sold
 
Dan, thanks for your post, I really appreciate the input.

Needless to say, I think we are all curious to know how this camera will perform.
 
Good to see you here Dan, I've been following your emails quite closely on the CML. Thanks for all the info, especially the recent bits on the tag 50740 metadata/dng headers. Interesting times!
 
Dan, thanks again for continuing to provide input here. If I'm not mistaken, RED cited a very serious lack of insertion/removal stamina as the reason for moving to custom interfaces. Apparently the drives are built to handle only a few dozen connections in their lifetime, which is perfectly reasonable - if not overly protective - for the average system disk which will get installed and then never touched again. As you alluded to, however, that expected number of lifetime insertions may be reached within a few days on a working set. It will be very interesting to see how they proceed. The idea of off-the-shelf prices is alluring, but if it can't be trusted, it's absolutely worthless.
 
In the pictures I see two blue e-sata ports, next to a USB port. I'm not sure what else those could be for other than accessing the SSD's mounted in the camera, so hopefully at the very least that will keep insertion/removal to a minimum. (for indies anyway. On anything more than a pretty small crew low budget film a DIT/data wrangler can't really do his job if to offload he has to steal the entire camera)
 
Fohdeesha said:
On anything more than a pretty small crew low budget film a DIT/data wrangler can't really do his job if to offload he has to steal the entire camera
Click to expand...


Er...yeah. If you can't shoot and offload at the same time, a lot of people could be pinched pretty hard by that. Or at least having the "luxury" of changing media when necessary, as in some remote documentary work.
 
Last edited:
Dan Hudgins said:
There may be an option to dump the SSD in one slot into a HDD in the other slot, then re-use the SSD.
Click to expand...

Kind of a band-aid solution, in my opinion, though. It doesn't really fix the problem - reliable, fast, removable storage. Cameras would still have mandatory downtime, except now people would be dealing with the added disadvantage of spinning disks...whose connectors are also prone to insertion fatigue.

I think SSDs with proper connectors are the best choice, but for ~$8000 and a desire to allow "off the shelf components", that just may not happen. We shall see!
 
Have you considered drive sleds with their own, more durable pass-through connectors? That way, you could fit SSDs into the drive sleds and use them without wearing out the connectors. That seems like it wouldn't be too horribly expensive to design and create, and would still let people use fast, non-proprietary storage devices.
 
Dan Hudgins said:
although so far I think they have done what few others could have.
Click to expand...

Agreed. Over the last five to seven years, a number of people talked about putting something on the table and "changing the game". Only a couple made it to the table, only one with wild success. Those at kinefinity.com (sm) have made me think it might be time for round two.
 
Darn, They beat me To it.... I was going to Reverse Engineer a red Epic , and Call it BLU... using China to Build it..... O well. saves me the headache
 
You must log in or register to reply here.
Back
Top