I've not had 1 single problem with my SDHC cards (I use the Panasonic ones) in about a year of usage with my HMC150. I'm quite happy with the codec as well (transcode to ProRes through NeoScene). It's not P2 or AVC-Intra, but still a damn good codec and Philip Bloom agrees on his blog about the AF100.
Panasonic reveals more details of AG-AF100
- Thread starter Quickbrownfox
- Start date
I've not had 1 single problem with my SDHC cards (I use the Panasonic ones) in about a year of usage with my HMC150. I'm quite happy with the codec as well (transcode to ProRes through NeoScene). It's not P2 or AVC-Intra, but still a damn good codec and Philip Bloom agrees on his blog about the AF100.
Barry_Green
Moderator
And a $17,500 price tag.
Not really, it's a digital cinema camera, not a video camera (it doesn't record video). But that's a whole different discussion.
$50 chip vs. $5,750 chip.
No, it's about the same as the AF100.
consumer division?
Barry_Green
Moderator
I don't believe there's a "magic number", there's all sorts of ways that they could do it, it all depends on what resources they can throw at the problem, at what budget.
The "magic number" for alias-and-moire-free 1080p from a single chip, could be as low as about 3 megapixels, not 12 or 14. 12 or 14 is just massive overkill. If you can read all of those pixels and use them, then it's not gonna hurt anything, but once you get beyond about 3 megapixels, it's overkill. The GH1 has 14 megapixels because of stills, not video. The Red has its massive number of megapixels because it does 3.2k, not just 1080p.
Which is why I hope that the AF100 will at least be doing some manner of binning, which will let them use the additional pixels for more sensitivity, rather than just for "bragging rights".
walnutcrunch
Well-known member
My impression from Panasonic is exactly this. They are building a 1080p HD video camera. Given the sensor they are using they are looking at it in terms of how do we use this sensor to create the best possible HD video. That's the difference in this camera. They are going for best possible HD video and nothing else.
Smart.
Otis Grapsas
Carbonite Member
The best results when skipping pixels are achieved at multiples of the resolution. All other cases are VERY inferior, they mess with luma and chroma in nemerous ways. For the megapixels of a modern dSLR that is 3840 pixels wide and 2160 pixels high crop. That would output a bayer image. Debayering such an bayer image is not an easy task though. If you use the standard methods it will artifact a lot compared to a true bayer sensor because the distances between pixels vary in the step of interpolation. If you design and use a new demosaicking method that takes this into acount it will produce better results. Actually, I have designed such a method
The wrong OLPF is not the main reason alias appears in dSLRs and that's easy to see on a bayer sensor without OLPF. It looks a lot better than a dSLR, so good that OLPF are not even used in many scientific applications that involve accurate measurements on geometry, pattern recognition, quality control in the industry etc. It's only a minor problem compared to what we see in dSLRs. A large part of the dSLR alias is due to the debayering algorithms and their difficult task in the case of pixel skipping.A smaller part is due to the OLPF filter. The current state of the art in developing like skipped bayer is not very good. Using a very heavy OLPF and sharpening after development can improve the look but it will look very soft compared to a true bayer sensor. So it will require very high sharpening after debayering.
If you can read all the sensor pixels at full rate, you can simply downsample with a good filter and it will look great, but that's not possible today. If you can only read all pixels horizontally (or vertically), that will reduce alias, but only on one axis. 5dII looks like they do avoid skipping pixels on one axis, so on one axis it has only OLPF alias, and on the other axis it has debayering artifacts AND OLPF alias (and looks far inferior).
On first image is the full bayer sensor. Setting the OLPF to a single pixel eliminates luma alias. The debayering methods are simple because there is tons of research on the subject in this case. All big companies have excellent methods to debayering a true bayer.
The second image is a camera that can only read 1/4 of the image. There is no ideal OLPF in this case between some green pixels are 1 pixel away and some are about 3 pixels away. If you set the OLPF to one pixel, it will alias luma horizontally and vertically. If you set the OLPF to three pixels, it will not alias but it will be soft. The development method will still have to be very good to produce a good image, even with a heavy OLPF. And the sharpening will have to be a very dedicated algorithm designed for this very task.
The third image can eliminate alias horizontally. It does not even need a different OLPF. The debayering problem still remains and alias will be heavy vertically. An improvement over the second image.
The fourth image is the same with the third image, but horizontally, not vertically.
