FS7: Shogun Inferno with FS7, Early Impressions

Thanks for the info regarding the limits on bitrate for UHD 60p footage. It will probably be pretty obvious once you get a chance to pull some 60p UHD material into your NLE and play with it. Being able to record in 60p was a big reason I bought this cam, so if it puts out UHD 60p at 10-bit I'd be pretty happy.

Regarding the "pseudo 10-bit" display on the Atomos, it uses an 8-bit plus 2-bit FRC dithering method to reach 10-bit, and this actually isn't necessarily bad. Eizo's brand new $2K 24" 1080p professional display (CG-247X) uses an 8 plus 2 FRC method to achieve 10-bits, and it's used to carefully scrutinize video and print work. So on a 7" camera monitor it *should* work just fine to reach a 10-bit color space.
 
basspig said:
I don't know what the bit depth is at 60P. I will do some experiments with shooting a gradient and looking for banding. If it's 8-bit, that should stand out pretty prominently. There is some question about bit depth and also color subsampling at this frame rate. It is, however full resolution and quality at 30P.
Click to expand...

We know from an earlier communication with Sony that the FS7 delivers via HDMI at UHD 50p/60p only 8bit. Seems to be due to the fact that UHD 50p has never been specified with a higher bit depth. That is why all the recorder are of lower interest, as long as one has to use HDMI from the FS7. And SDI from the FS7 is limited to HD too.

The only solution would be to use the raw output, but especially that is limited with the Shogun Inferno to UHD 30p.
 
For long form concert recordings, I think we can get by with 60P UHD. At any rate, the data is enormous.. a pan across my studio resulted in a 4GB+ file! Imagine a 95 minute classical concert x 4 cameras. I may limit that trick to one stationary camera that I can apply panning and zooming to in post.


If Atomos uses the same technology for creating 10-bit display quality, then the problem must be in the Atom HDR algorithm. Of course, when I turn that off, the shadow areas lighten well past the "magenta bloom" threshold. So native mode may be masking the problem with the S-Log curve. The video, though shot off the monitor, does show this pretty clearly.
 
I agree that 10-bit UHD at 60p might be overkill for something that runs for a hour or more (since you can simply expose for it correctly, and 8-bit shouldn't matter), but I'd like to be able to shoot a full day of, say, event coverage at 60p UHD 10-bit onto a couple 1TB drives and make 1 extra swap of media instead of (9) 128GB XQD cards.

I checked out your video Mark. Thanks for uploading that. Personally, I think I might prefer the magenta in those really deep shadows (since it doesn't affect the recording) because it would work like a false color that lets me know where the cutoff between true black and deep black falls. Just based on what I saw anyway...
 
I'd rather not have the 'false color' effect in shadows like that, unless I have that feature enabled. It's real annoying and it accentuates any noise down there as well.

Shooting anything long form that's not MPEG HD is a massive disc hog. I'm already pondering upgrading all the SSDs in my NLE to the 4TB Samsungs. But I feel more pressure to upgrade the CPUs to the faster 3.2 GHz Xeons.
 
Just an observation: set up the FS7 on a white surface with soft lights to create very moderate gradients.

Observed the Shogun Inferno display both at 23.98P and at 60P (UHD) and noted the banding appeared the same for both frame rates. I was actually surprised to see that much banding on a supposedly 10-bit display. Not to say that it isn't an excellent display, but if 60P out of the FS7 is 8-bit and 24P is 10-bit, I sure as heck can't tell the difference on the Inferno's display.

I also turned on the maintenance menu SAW pattern and noted banding with some false colors on the Inferno display.

As for the false color/posterizing near shadows when in Atom HDR mode, that problem exists to a lesser degree in the Log to Video and Custom mode. It's completely free of it in Native Source mode. The problem with this posterizing is that it greatly exaggerates noise in the picture, where other modes look completely clean, to say nothing of the fact that it mispresents the color of dark objects as magenta.

If one can avoid the scenes where this is a problem, Atom HDR is really quite amazing.
 
Interesting observations. I've had my eye on the Ninja Flame because being able to shoot in sunlight is a big deal to me (the log mode would also be interesting).

But Mark, you say that the ProRes 4K actually looks better than the 10-bit X-AVC I that the camera records internally? If so, that might really convince me.
 
Barry, the display is simply amazing for outdoor, direct sunlight use. I've never seen a display that I could view easily in bright conditions. For that feature alone, it might be worth having.

Prores HQ is far superior to the XAVC internal recording. I noted two things different:

At 400% magnification (pixel peeping), the XAVC shows prominent block artifacts. Fine details are slightly muted too.

At normal viewing on a DCI 4K display, a scene I shot with dense foliage against blue sky taken with polarizing filter and using a LUT that gave me horrid artifacts on the XAVC version, there are no artifacts to be seen. Thin branches some 500' away are clearly rendered to the pixel (35mm prime lens at f/2.8), yet the blue sky has no horizontal magenta lines through it, as is the case with the XAVC version. The data is much heavier, but if the destination is going to be the big screen, then it's worth it. Of course, most of this is lost the minute it gets rendered to a distribution CODEC. But the master footage looks breathtaking.

If I were to rate which CODECs have similar quality, I'd say the Prores 422 is similar in quality to XAVC, showing some block artifacts. HQ is significantly better than 422.

My observations are based on S-Log3 S-gamut with custom LUTs. Blu sky artifacts are largely affected by the LUT used. Block artifacts in dense foliage are observable with any LUT.
 
Hi Mark!

I'm getting more and more excited with the Inferno, basing on your posts. Could you please also check for us how DNxHR is doing, compared to XAVC? I wonder if it will be as good as Prores. TIA

Piotr
 
Wow, somehow I missed the memo on ProResHQ doing a better job than the internal X-AVC. Or maybe not too many people care about that last bit of quality... but I'm impressed. It makes me wonder if the Video Devices PIX-E5/7 that records in ProRes 4444 XQ 12-bit is able to pull any more quality out of the HDMI signal, or if ProResHQ already fully takes advantage of it? It's probably very small if noticible at all. Thanks for the info!
 
Piotr,

The DNxHR seems to be about equal quality, but is limited to video legal levels, hence it produces lower contrast images than the Prores.


Barry,

XAVC certainly is a fine CODEC, given the amount of compression it employs, but some of us have found certain shooting conditions where that algorithm breaks with rather ugly results. I noticed it in September, when shooting through trees with a polarizing filter. Lots of foliage filling the frame, all in motion. Flat blue sky in between. And magenta lines running through the blue, as well as gross pixelation of the edges of a cloud in that sky. I thought my camera was broken! Then I shared it with others here and I think Piotr also found a similar problem shooting similar scenery and posted about it some months back.

The weak areas are in shooting anything with lots of high frequency detail, especially if there is motion. I recall a particularly nasty artifact shooting a waterfall through a wooded foreground. There were more woods surrounding the waterfall, too. Lots of HF detail, plus rapid motion of the waterfall itself. Sometimes you could see block artifacts at normal 1:1 view magnification.

It's easy enough to compare the quality. Some subjects that break CODECs are:

Dense foliage
Fire, zoomed in close so it fills the frame
Sunlight Specular highlights on choppy ocean surface

I just shoot a static scene of the woods in my back yard and compare the XAVC to the Prores HQ. At 400% magnification, macroblock like artifacts are visible in the XAVC, but not in the Prores HQ. If you use the intermediate Prores 422, you'll also see some macroblock like artifacts. I can't speak for the 4444 XQ mode. It may be wasted on 422 source material.
 
basspig said:
Piotr,

The DNxHR seems to be about equal quality, but is limited to video legal levels, hence it produces lower contrast images than the Prores.
Click to expand...

Would you care to tell me which exactly flavor of DNxHR is Inferno recording? In Resolve, I'm using the HQX 444 version, which definitely is capable of data and not just video levels... I was hoping I could record in the same codec! Thanks,

Piotr
 
Mine came in after waiting since it was announced. Tested it in the shop and found it to be defective. Guess I'll have to wait a few more weeks for another one to come in.
 
Ouch! Quality control isn't that great.. I have four Atomos products, dating back to the original Samurai. All have been 100% reliable so far. Sorry to hear you had gotten the bad apple.
 
What are the other features besides RAW support (CinemaDNG) that they promised? Just wondering...
 
You must log in or register to reply here.
Back
Top