Graeme Nattress

"from the fact that they're really rendering about a 2K image (less, with antialias filtering), blowing it up and calling it 4K." Last year you claimed it was a 1k image Phil. At least make your mind up about which anti-Bayer CFA FUD you want to go on about. It amazes me, that in one post you can describe, in language anyone can understand all about FPGA and ASICs and their programming, and in the next go so extremely off the planet about Bayer CFA stuff.

The mac I'm working on now has been up and running for weeks. I code on it, I watch movies, email, web, etc. etc. Only reason it's weeks, not months is that it wasn't been used over xmas so it got switched off.... You were asking what I do at RED? Primarily image processing / codec development, but I'm involved in many, many aspects of the camera.

Carl, http://scien.stanford.edu/class/psych221/p...ngchen/main.htm will probably give you a good idea of how it all works. Traditionally, consumer cameras have used over-simple algorithms. We can use much better algorithms because we have the horse power in camera and in computer. There's no reason why consumer cameras today couldn't do the same.

Answer is: very representitive. If you stick a high quality algorithm on the reconstruction, you get really, really nice, images. Perhaps you'd better come over to reduser.net though if we're not allowed to talk about RED here? Graeme

www.reduser.net is the site if you're interested. I don't know what you mean about making it more of a 2k camera. I'd say current changes are exactly the opposite of that. Graeme

No reason. That's just the way the graph was drawn by excell. Graeme

Because that was the point where we couldn't discern between that and the next wedge - ie, the noise was stronger than the signal. Graeme

Be going universal on the plugins in the new year. Given that the D20 is a 12bit A-to-D, it's curve appeared almost correct for what I'd expect. It appeared to have a little more range than I'd expect from it, but it looked reasonable. I don't think our new test is biassed towards RED, more than biassed away from potential sources of error. It will be interesting to run some more cameras through it, but we'll have the same issues that Geoff has that video cameras don't come set for best dynamic range out of the bag. Graeme

If you click through you get links to upsampled DVDs with native HD frames. What kills me is that there's still excessive edge enhancement and ringing artifacts on the HD stuff. Ouch. The HD still also looks soft. Graeme

It all depends on what the image processing did in camera to the highlights and what you define as when the noise swamps shadow detail. That can be very subjective unless you adopt a better testing methodology. Graeme

Yes Phil! That's totally right. And without any specified repeatable metric for acceptable shadow noise, where do you place black clip?? Graeme

Stephen - Exactly. Geoff and David haven't looked at the RAW camera data, but "processed" data. I don't know what they did with my software, or what settings were used to generate the images they used. As you say, other cameras would have performed better under different conditions and I agree. We have indeed improved the sensor's performance since David used it. Graeme

Well, you've got 6 cameras, tested at 6 different times, with different procedures. That chart is about as useful as a chocolate teapot. Not least it's comparing finished, in the field cameras, when RED is still in heavy R&D and does not yet exist as a fully finished camera. Indeed, the tests that David did with us did allow us to change some sensor variables and for us to achieve better results. Without a proper test methodology, such tests, when combined together like that, reflect more on the procedure itself, rather than the devices under test. Graeme

RED is a 12mp sensor, BTW. And yes, the problem is getting such a large and high resolution sensor running at 24, never mind 60fps. Graeme

We were just talking about the command line version today. Can't say for definate, but we're very, very seriously looking into it. If you want to email me off-board, I'd be keen to hear your throughts of what you'd like in the app and how you'd use it. Graeme

Nate, be interesting to hear from you when you get to try the Pentax. One thing I noticed is ISO only up to 1600 - if they were as noise free as to warrant 22bit, I'd have expected a higher max ISO rating? Anyway, it's all spec-ulation until we get to play with one and see how it performs. Thanks, Graeme

And if that camera had a 132db SNR, I'd believe you! If it's downconverts to 12bit, which I would be a reasonable figure for such a CCD camera, it would have a SNR in the range 66 to 72db, which is quite reasonable. To quote DPReview - "How relevant this '22 bit ADC' is to the final image quality is yet to be proven." - ie, unless this camera is essentially noiseless under practically all circumstances, which is what a 132db SNR would imply, that 22bit claim is utterly pointless marketspeak. Graeme

Jim is monitoring dvxuser.com, dvinfo.net and working directly with the likes of Warner Bros., David Fincher, Peter Jackson, David Stump and Otto Nemenz for direction and feedback. There is no "official" RED board. One of the nice things about taking feedback from the raw conversion, is that we can show how RAW works - so much nicer to decide how to process that image in post, rather than make a mistake in camera and burn in a certain look. Graeme

Chris is right about active development (and late nights). We're still doing a lot of fine tuning on the RAW conversion software, hence the much better image Jim posted after people said "um - nice, but what's up with... " etc. on the original. We're very happy to take constructive criticism, but people have got to remember it's a project in development. Sure it would be lovely for us to give out RAW 12bit images, but it's just not practical for us to do that, for many reasons. Similarly, no other camera company gives even the level of early imagery that we've managed to get out. Usually, the first time you see moving images is when the camera is complete and it's in the hands of a reviewer. So, yes the original RAW data is 12bit. But I must take to task the comment about a 22bit DLSR. I'm sorry, but it doesn't exist. The best medium format backs still only give out 14bit data (probably embedded in 16bits). Graeme

Digibeta is significantly better than DV. It's about 2:1 compression, rather than 5:1 and is certainly better than visually lossless over multiple generations. Digibeta is 10bit quantisation, 4:2:2 chroma sampling compared to DV which is 8bit and 4:1:1 (or 4:2:0 in PAL land). Graeme

There's a 2k rez version up on the RED site now. http://red.com/images/gallery/still_5-downsize-for-web.jpg Graeme

If you open up the JPEG in photoshop and examine the values of the regions you describe as being clipped, I think you'll find that a) they're not clipped and B) by adjusting the black level, you'll find that they're still holding detail.

Thanks Jason! Graeme

Please don't jab at a workflow you don't understand, Jason. It's not clever and it diminishes the excellent news you have to tell people about Iridas.

We'll make as many as we can sell. I don't think there's any legal reasons why we'd only make a small number. The reservation scheme was just accurately judge manfacturing volumes. Why are we stopping reservations? Because we have more than enough reservations to know we'll get the volume we need.