[k5astro]

Okie, a 'list'... of video specifications/features/changes that would turn K3, K5 and K-01 in to AWESOME Video Production Tools.

01. 'Raw' colour space profile - full sensor data, not limited to REC709.
02. Clean Feed of the raw sensor colour space to the HDMI 4.2.2. output, 4.4.4. if possible, sent to the HDMI port before the downscaling to the LCD is done.
03. Implement optional CoDecs - h.264, high bite rate Mjpeg, and Compressed CinemaDNG Raw - with h.264 and Mjpeg having user selectable bit-rates.
04. User selectable removal of all the post-production applets (to free up system space for the CoDecs)
05. 48kHz sample rate Audio. (with the lowest damn sound floor setting possible!)
06. User selectable ISO levels - enable all the 'hidden' ISO's that the camera can use in Auto so that the User can choose them when needed (ie, 80, 120, 240, 320, 640 etc etc )
07. No Electronic SR. - Mechanical SR on the Sensor or in the Lens.
08. No 'extra' crop between Stills and Video modes when SR is OFF.
09. Full manual control of ISO, F-stop, shutter speed during recording for any camera that doesn't yet have it.
10. USB Streaming - with activation automatically limiting recording to h.264.

On point 04,. I'd be happy to see a 'pro' firmware for those who want the high-end features listed above, and a 'enthusiast' level firmware that includes auto-everything modes and the tool applets.
They could call them "vP1.0x" and "vE1.0x"

Thanks for the list PiDicus! What do you think would be the minimum required for the semipro? For example, If someone were to actually make a documentary with they would probably use external sound equipment, right? And USB streaming is a "nice-to-have" I would think. I looked at Final Cut and Premier and neither support CinemaDNG. Why not just h.264?

Can the rolling shutter be improved? (Even the Red has rolling shutter.)

---

Last edited by k5astro; 02-02-2014 at 07:37 AM.

[kadajawi]

Thanks for the list PiDicus! What do you think would be the minimum required for the semipro? For example, If someone were to actually make a documentary with they would probably use external sound equipment, right? And USB streaming is a "nice-to-have" I would think. I looked at Final Cut and Premier and neither support CinemaDNG. Why not just h.264?

Can the rolling shutter be improved? (Even the Red has rolling shutter.)

USB steaming could be nice if that means you can stream to a SSD when shooting Cinema DNG. The amount of data, and the speed needed could be a bit hard on SD cards. I wouldn't see it as a big priority though.


Cinema DNG... it requires a different workflow, yes. But the advantages can be huge, and it might even be the easiest to implement. The K-3 can save DNG files. All that is needed is low enough resolution DNG files at fast enough frame rates, plus I don't know what magic they do to sync it to audio. It may be relatively "trivial". Any other format like ProRes would probably require significantly more work from Pentax, if it can be done at all. h264 in the current iteration is... well, you can work with it, but there are limitations. Say if you film a walk through a forest, with the lights coming through between the leaves, and lots of other details, and everything is in motion, then you'll see blocking. Dark areas may also fall apart relatively soon (details go away, lots of big ugly blocks and banding), as that is what h264 likes to throw away first (you wouldn't see it anyway). Problem is if you try to grade the video, lift it up a bit perhaps, you will certainly see that.


As for advantages... I think it was mentioned before, Cinema DNG is simply DNG, while h264 is highly compressed JPEG. Try edit such a JPEG vs a DNG file. You wouldn't have to get the white balance right in camera. You have a bigger dynamic range to play with. etc.


There is h264 that only saves intraframes, which makes it more suitable for editing and won't have degrading image quality between I frames (though a good encoder wouldn't have that problem). Basically it's MJPEG, just with the more efficient algorithm for saving stills that h264 has (JPEG is quite dated...). Pentax could also try to up the maximum bitrate (it would have to with intraframes) and the quality settings. Not sure if that can be done... I have the feeling that the hardware encoder isn't too flexible (there aren't even quality settings... like 1 star, 2 stars, 3 stars). There is also 10 and 12 bit h264 (vs 8 bit that is commonly used everywhere). Playback can be a problem (no commercially available hardware decoder exists that can play it back with hardware acceleration, graphics cards also won't do it), and I really don't think the processor in the K-3 is flexible enough for that.


There used to be a Sigma DSLR that could only shoot RAW files. No JPEG. I'd think it was because Sigma didn't have the technology to (competitively) convert the sensor data to good looking JPEG files. To get white balance etc. right. "Just" saving the RAW files straight to the SD card might simply be the easiest way to do it.


@gbeaton: I doubt it. Not in firmware. I could imagine that they could save lots of accelerometer and gyroscope data, as finely grained as possible, and also detailed data about the shutter speed etc., and record in a higher resolution than needed, and then write a program that analyses the data and adjusts the position of each line accordingly. You'd still end up with a crop that you couldn't frame for, but this might eliminate rolling shutter. And it is absolutely not going to happen.


However the most noticeable negative effect of rolling shutter is jello. The result of small, fast movements in all directions, randomly. You'll easily see it, and it is really distracting to me. THIS can be fixed by stabilizing the video... even if you are going for a handheld look, these jerks are probably not what you want. There are 2 ways: Stabilize the video in the camera or lens. The K-5 did it, the OM-D does it, and when using a lens with IS on other brands it also works. Or you stabilize the camera by using a rig (and learning how to use it). Once you've got rid of this RS becomes a small issue IMHO. Fast pans will show it, but those don't occur that often. Try to avoid it, or use some Premiere Pro function or plugin to fix it. Since it is a much more constant movement it's probably easier. Or just ignore it, chances are viewers won't notice. I've seen blockbusters with rolling shutter issues during fast pans... doesn't seem to be that problematic to viewers.


There is one feature I'd like to see on that list: WB-L. Like AE-L, but using the WB button during video shoots. Since we need to get WB right in the camera, and we may need to get it not to constantly adjust this could be quite helpful. The auto WB often gets quite close to what is correct, but it's deadly if it is shifting around.

[kenyee]

@kenyee: It does overscan, but in the camera. I don't think it can be done, what you suggest. The encoding hardware would have to be quite flexible...

Most encoding hardware actually is flexible. Main issue is the H.264 format doesn't have defined profiles to put the overscanned data into.

I'd also wish the K-3II does 1080/60p so we can do slow motion but the 3rd party chip they use doesn't support it so it isn't currently possible

[PiDicus Rex]

I wrote in USB Streaming for all the people who do 'live-to-online' work, like is done for online events - it's something that is becoming more common.
It's not something that would be done using anything more then h.264, which is the preferred Codec for online delivery.

To record to an SSD, we'd all be better off with the 'clean feed HDMI' and using an external recorder like an Atomos.

h.264 is a 'content delivery' CoDec, not a 'production' CoDec - it's very good for streaming and packaging content down to exceptionally small sizes for transmission, but it throws away a heck of a lot of information, especially fine detail.

If you ever get to see h.264 side by side with the ProRes recording made via HDMI out from the same camera that's recording the h.264, you will be stunned by just how much difference there is.

I need to correct one thing here,..

As for advantages... I think it was mentioned before, Cinema DNG is simply DNG, while h264 is highly compressed JPEG. Try edit such a JPEG vs a DNG file. You wouldn't have to get the white balance right in camera. You have a bigger dynamic range to play with. etc.

h.264 is Long Group Of Pictures Mpeg4 ( Long-GOP ) it is Inter-frame compression that occurs over many frames of video.
MJPEG is Intra-frame compression, where each frame is compressed using Jpeg algorithms.
Compressed CinemaDNG is assembled in a similar manner to MJPEG, only the wrapper is a DNG still for each frame, rather then an AVI or MOV file.

You'd still make sure the White Balance is set properly in the camera - that's just good operational procedure - but yes, DNG lets you adjust nearly Everything in post production - Colour Temp, Exposure, balance between colour channels, gamma curves and more.

The difference in the math behind the two forms of compression is how they deal with changes that occur within the frames across multiple frames.
It might make sense to explain that one is a Jpeg compression, where the other is a Temporal Jpeg compression.

i-frame only h.264 still throws a lot of information away, but does have a higher data rate - 45mBits/second on the 5D3, compared to about 20 for the ibp profile - it's better, but still looks nowhere near as good as ProRes or CineDNG

h.264 also puts a lot of load on the Editing system resources - all that extra compression for delivery profiles means there is a lot more math done to open and scrub through the files then other formats - A Pentium 4 can be used to edit DV or HDV but feed it Mpeg4 and the PC will nearly grind to a halt.
Even recent PC's will struggle with h.264 when scrubbing backwards or editing multiple layers.

ProRes has more detail, and because it is Less compressed, it's easier to edit.

Compressed CinemaDNG is a sequence of Tiff's, one for each frame of video, with each Tiff separately compressed using Jpeg processing. It is substantially more data to shift around, but the actual rendering requires less CPU power, assuming you don't go overboard with S/Fx, because there is less compression math occurring.
And because the frames are compressed separately, you don't get blocky break up that travels across multiple frames.

There is however, a heck of a lot more data being shifted from hard drive to memory to CPU and out to the GPU, so you still need a high end PC to do it.

This brings me to the last limitation, the speed of the SD Card and the controller, and how much data can be shifted through it is the final bottleneck on using higher-end CoDecs.

So we're all fortunate that the K3 has a UHS-1 controller.

PS - I edited this comment a heap as I tried to make it as clear as possible - CoDecs are one of the most complex parts of video production, and are not easy to get a grip with.

---

Last edited by PiDicus Rex; 02-03-2014 at 03:26 AM.

[kadajawi]

If you want to do online work (say, streaming), the clean HDMI out is more important. Add a TriCaster or similar and voila, you can stream.

Intraframe is possible with h264. However I do think that the image quality, at the same bitrate, will be better with interframe compression. You have more bits available for moments where you need them, for example if you have a very detail rich scene that needs huge JPEG files, but where nothing moves. Intraframe has the biggest advantage with editing, or when the interframe encoder does a bad job. I tend to encode my Blu Rays with very, very long I frame interval, sometimes as long as 40-60 seconds. They look perfect though, just skipping through the video is... not recommended. (The Dark Knight Rises, with 2 surround audio tracks, ended up being 5.88 GB. You barely notice the difference to the original BR, there is no banding and hardly any banding, it's mostly just a little bit less noisy).

Compressed CinemaDNG would be easiest on the SD card (MJPEG levels of bitrate I suppose? Maybe a bit higher), however the K-3 does not support that sort of lossy compressed DNG. Regular DNG, whether compressed or not is still a TIFF (more or less). It is lossless. And that's what the K-3 can save. The main question is if it can compress/save at least 24 2 MP DNGs a second.

[PiDicus Rex]

The really advantageous thing, when we're discussing the K3 and CoDecs, is that it already has h.264, Mjpeg, and DNG built in.

One for Video.
One for Interval Movies.
One for Stills.

It's all there, we just need it applied in a different manner, and it'd be nice if it was the same on K3, K5 and K-01.

[PiDicus Rex]

The main question is if it can compress/save at least 24 2 MP DNGs a second.

The answer to that, is related to how many 24 megapixel raw stills it can save in burst mode,.....

EDIT: I read the manual, specs show 8.3 frames per second.
8.3 still frames x 24 megapixels is 199.2 million pixels per second.
24 still frames x 2.2 megapixels is 52.8 million pixels per second.

I'm thinking Yes.

( 60 stills at 2.2 megapixels is 132 million pixels per second. . . . . )

---

Last edited by PiDicus Rex; 02-03-2014 at 03:46 AM.

[kadajawi]

The answer to that, is related to how many 24 megapixel raw stills it can save in burst mode,.....

EDIT: I read the manual, specs show 8.3 frames per second.
8.3 still frames x 24 megapixels is 199.2 million pixels per second.
24 still frames x 2.2 megapixels is 52.8 million pixels per second.

I'm thinking Yes.

( 60 stills at 2.2 megapixels is 132 million pixels per second. . . . . )

Oh but it can't. It can flop the mirror fast enough, it can read out the sensor fast enough. But there may be a huge buffer in there, and the files are only processed afterwards. The question would rather be what is the sustained framerate the camera has once the buffer is full. Basically how fast can it write data given a very fast SD card, and what is the speed of the processor/how fast the compression happens. It might be that it is not the write speed onto the SD card that doesn't allow for sustained high speeds, but the speed at which the processor compresses the files. Then there is the question how well can the picture be compressed. A 24 MP DNG is probably smaller than 2 12 MP DNGs of the same subject (unless we are going for completely uncompressed files... in which case the processor speed should be less of an issue). Not clue if there will be speed differences between many small files and few big files.

So basically... there are quite a few variables, I personally think it can be done, some way or another (perhaps uncompressed (TIFF) DNG files saved onto alternate SD cards, and without audio, but that would be a solution that would be so consumer unfriendly they may not be willing to do it...).

Btw., if it were possible, I'd like them to read out more pixels that are then downsampled before saving. If the whole sensor is too much, maybe instead of every 12th pixel or so make it every 6th pixel. Whatever can be done and makes sense (would it be better to spread out the read pixels evenly, basically taking a 4k video and downsampling it, or would neighbourings pixels make more sense, ...).

But speculating on this is rather futile, we don't have any knowledge about what the sensor can or can not do, and likewise the processor.

I think at this point we can completely forget about the K-5. That's history, it wouldn't make any sense for Pentax to do any work on it. Same with the K-01. The K-50 and K-500 are out too, cause they are consumer focused cameras. It makes sense to give the current top of the line camera these functions. Or to even split off the K-3. Make a K-3v... K-3 body (maybe different labeling on the buttons, or maybe even an all new body that drops a few still parts (mirror) and puts the internals into a metal box with mounting points for all sorts of handles, grips, ..), K-3 internals (maybe upgraded), but special firmware that adds the extra features. The price could be raised as a result, say by a few hundred to a thousand, depending on how good it is.

[PiDicus Rex]

Imma needs time to think, you raise some good points, but I still feel it is the data throughput that will determine the write-speed limitations, and consequently, which formats can be accomplished 'in camera'.

This is also why I keep bangin' on about Clean Feed HDMI Output (in 422 or 444), as it would mean we could start using external recorders that are built to record in these formats from the get-go.

[kadajawi]

Imma needs time to think, you raise some good points, but I still feel it is the data throughput that will determine the write-speed limitations, and consequently, which formats can be accomplished 'in camera'.

This is also why I keep bangin' on about Clean Feed HDMI Output (in 422 or 444), as it would mean we could start using external recorders that are built to record in these formats from the get-go.

But isn't the only way to get actual RAW files to get access to the sensor? Meaning you can do stuff like ProRes with those recorders, but actual CineDNG, where you are free to chose the white balance etc... I don't think so.

[Steve.Ledger]

What has the K-3 operator done wrong here?

You Tube

You Tube

[kadajawi]

No or wrong deinterlacing? I have the feeling he used the wrong line order when deinterlacing (or whatever that is called... as in odd lines first instead of even lines first).

[k5astro]

What has the K-3 operator done wrong here?
Pentax K-3 video III. (1920 x 1080, 50i, MOV) - YouTube

I don't know but the picture is rocking a little... and it doesn't look so good even at 720p.

[Steve.Ledger]

Full of really awful aliasing from skipping too many lines by the looks. Like it was shot at 720p and upscaled in post to 1080p.

[kadajawi]

The title says 50i though. Maybe he uploaded the original file, without deinterlacing it first, and YouTube screwed up the deinterlacing/didn't deinterlace it at all, but tried to compress it to a standard (low) YouTube bitrate...? Interlaced footage, when not recognized as interlaced, should require pretty high bitrates...