Sony FS5 showing Artefacts in UHD mode is “NORMAL” Update

Categories: Miscellaneous 12 Comments

pxwfs5_front_8bit

Many Sony FS5 users are experiencing artefacts in UHD mode which they seem to think is a processing problem but 3200 ISO with standard gamma is +9dB gain so its noisy and on top of the 8 bit codec its compressed 32:1…you can’t avoid artefacts.

Alister Chapman ” I believe the edge tearing artefact reported by some is due to the temporal noise reduction that the camera uses. As you pan or tilt the camera the areas around edges will change from frame to frame while large more solid areas of the picture change little. Because of the differences from frame to frame you won’t get any noise reduction in these areas so edges can become coarse and “fuzzy” or jittery as there is on the one hand a marked increase in noise on the edges and at the same time a ghost image from the adjacent frames. At 0db this is barely visible and certainly not an issue with real world shots but ramp up the gain and this increases rapidly. These are fairly typical side effects of temporal NR.”

“The FS5 is a great little camera, I really enjoy shooting with mine and I think the results I am getting are great. I know that I can get a technically better 4K/UHD image from my FS7 or F5, but sometimes it’s not just about getting the best technical quality. The FS5 allows me to shoot more freely, it’s a breeze to carry around or travel with, I can throw it on a gimbal, on the end of a microphone boom pole, chuck it and a bunch of lenses in a small back-pack, it’s fun to use! As a result I’m getting shots that I just can’t get with the FS7 or F5.”

See more about this issue at XDCAM-USER  http://www.xdcam-user.com/2016/01/my-response-to-all-the-fs5-artefact-issue-complainers/


Andrew v2

Andrew Reid from EOSHD counter claims Alister Chapman and reckons the edge tearing is a bug that Sony need to fix, Andrew also owns the Sony FS5.

You can read Andrews response here http://www.eoshd.com/2016/01/response-alister-chapmans-response-sony-fs5-complainers/

HDW : I will be running my own tests today as I am probably one of the few not to rush in and update my FS5 to firmware 1.10 though there is no indication that the new FW is to blame.

Sony seriously need to address this ASAP by making a statement via Japan from the people who know this camera inside out, I do feel Alister’s +9dB scenario a bit hard to grasp as I have always tested most of my review cameras up to 9dBs and the majority of large sensor cameras cope well at 9dBs, especially super 35mm sensors.

It does not help that the “Native” ISO changes dependent on wither you are in standard gamma or Slog !

 

Confuse v2

Here is a video which clearly shows the 8 bit UHD compression artefacts…

Sony FS5 Artifacts in UHD from Glass Ink Media on Vimeo.

For all your video production needs in Scotland, get in touch with Small Video Productions

12 comments on this post

  1. Max says:

    This is incorrect. The issue in the posted video is from the camera’s temporal noise reduction. So it is in fact an image processing problem. Nothing to do with codec or 8 bit, etc. Alister Chapman acknowledged it in his latest post.

    “I believe the edge tearing artefact reported by some is due to the temporal noise reduction that the camera uses. As you pan or tilt the camera the areas around edges will change from frame to frame while large more solid areas of the picture change little. Because of the differences from frame to frame you won’t get any noise reduction in these areas so edges can become coarse and “fuzzy” or jittery as there is on the one hand a marked increase in noise on the edges and at the same time a ghost image from the adjacent frames.”

    HDW : As usual you choose to edit out his last comment to suit your case, I on the other hand have printed Alister’s statement in full.

  2. Max says:

    The only case I’m making is that the artifacts in the video embedded on your site are not artifacts from 8-bit UHD compression. Your post is inaccurate. The artifacts are the same on the uncompressed HDMI feed. Sony tech has already confirmed with me that it’s not a codec issue. Alister goes on to address complaints made by other people about noise on the FS5, but the noise reduction tearing shown in that video has nothing to do with that so it’s irrelevant.

  3. David Heath says:

    I’m confused by the comment “…but 3200 ISO with standard gamma is +9dB gain so its noisy …” – I’d thought ISO3200 on the FS5 WAS it’s base ISO? If it’s equivalent to +9dB of gain, that makes the base ISO 1.5 stops less – about ISO1250? Yes?

    Apart from that I am struggling to see a problem on the embedded video. That’s not to say there isn’t any – but “clearly shows the 8 bit UHD compression artefacts…”???

    Whatever “problem” may be there, I think it’s important to get a sense of perspective. What is being talked about is material gathered at ISO3200. Whatever the base ISO of the FS5 is or is not, I’m used to a typical video camera ISO (eg of an EX1) being of the order of around 400? Which is itself considered pretty good – over a stop better than comparable 1/3″ cameras.

    In other words, what is being shown is equivalent to an EX1 with about +24dB of gain in!!! Does anybody seriously think an EX1 with +24dB would look even remotely as clean as the “test” above!??! If you want ISO3200, it’s frankly ridiculous to expect no “artefacts” at all. The video above may have imperfections – but it’s still one of the best results overall I’ve seen at such high ISO ratings.

    Finally, I have to agree with Max in that whatever is being seen, it is not going to be anything to do with 8 bit. Codec and compression? Conceivable – but bearing in mind what Sony and Alister seem to be saying then I’m minded to agree inter-frame noise reduction a more likely culprit.

    In which case, users have a choice. Accept as is – or go for less noise reduction. It’s the same as with the Panasonic HPX370 a few years ago – though with the FS5 far less severe and at far higher ISOs.

    The choice of differing modes nowadays on cameras MAY give a lot more flexibility – or it may just mean users getting confused and using settings inappropriately. Classic example may be using s-log AND gain at the same time! As Alister puts it: “….please understand that S-log only does one thing well and that’s maximise dynamic range. If you can’t get a correct exposure or better still a 1 to 2 stop over exposed exposure at the native ISO you really, really shouldn’t be using log. It will be noisy, it will be grainy, it won’t look good ….” Yet I wonder how many people do use s-log with gain…..?

  4. Chris Young says:

    David. The base ISO is considered to be 1000 but this can change depending on the gamma curve selected. If you download the actual 1080 file from Vimeo and have a good look at it the problem is very obvious and not something I have seen before. It needs to be addressed by Sony and I have no doubt it will be in further firmware upgrades. So far Sony have fixed quite a few issues with the F5 and F55 and I don’t really see why the FS5 should fare any differently.

    If interested these are the various ISOs available on the FS5

    The value you can set depends on the
    [GAMMA] setting of the picture profile.
    – STANDARD: ISO1000 to ISO25600
    – STILL: ISO800 to ISO20000
    – CINE1: ISO800 to ISO20000
    – CINE2: ISO640 to ISO16000
    – CINE3: ISO1000 to ISO25600
    – CINE4: ISO1000 to ISO25600
    – ITU709: ISO1000 to ISO25600
    – ITU709(800%): ISO3200 to ISO80000
    – S-LOG2: ISO3200 to ISO80000
    – S-LOG3: ISO3200 to ISO80000

    The value you can set depends on the
    [GAMMA] setting of the picture profile.
    – STANDARD: ISO1000 to ISO25600
    – STILL: ISO800 to ISO20000
    – CINE1: ISO800 to ISO20000
    – CINE2: ISO640 to ISO16000
    – CINE3: ISO1000 to ISO25600
    – CINE4: ISO1000 to ISO25600
    – ITU709: ISO1000 to ISO25600
    – ITU709(800%): ISO3200 to ISO80000
    – S-LOG2: ISO3200 to ISO80000
    – S-LOG3: ISO3200 to ISO80000

  5. Cliff says:

    Thanks Philip for getting involved in this important discussion!

    Could you please shoot some UHD on resolution charts? We know that this problem is worse with straight lines and mild camera movement. Sometimes “thick” lines looks worse too.

    As far as noise reduction goes, we all know what noise is and when we see it with gain. We also know what GOOD noise reduction looks like. What it seems that we are seeing with the FS5 is BAD or WRONG noise reduction being applied.

    +9db of gain? That is NOTHING for Sony’s excellent noise reduction algorithms today! That is, IF it’s properly tuned and applied. I think the FS5 is NOT at noisy camera at +9db, I think that it’s noise reduction processing is clearly DEFECTIVE. Hopefully it’s an easy Sony firmware fix to return it to where Sony usually is with it’s algorithms today.

    Thank you Philip and I’m glad to see you jump into the crowd with this issue.

  6. David Heath says:

    To Chris Young – I am still unhappy about where the “base ISO” figures are coming from. Forgive me if I misinterpret, but if the assumption is that it’s because ISO 1000 is the lowest it can normally be set to, then that is not necessarily correct.

    Reasoning is that the “least sensitive” setting does not necessarily correspond to 0dB. We’re all used to standard ENG cameras which allow for negative gain. So maybe the minimum setting corresponds to -6dB? Maybe even more negative? So if ISO 1000 was the same as -6dB, 0dB will be ISO 2000?

  7. David Heath says:

    Having now read Andrew Reids comments, then whilst I soundly agree with one point he makes in a very practical sense (which I’ll come back to at the end!) I remain unconvinced by many of the technical points he is making.

    Firstly, he originally put down the cause of the artefacting to 8 bit and codec/compression. The evidence now seems to be far more strongly that it’s as Alister says – inter-frame noise reduction. (And Andrew himself seems to be accepting that theory is more likely now?)

    Secondly, then on the subject of s-log then I also go broadly along with what Alister says about dynamic range and not using gain. You have to think what is happening AT SENSOR LEVEL. One end of the dynamic range is fixed by the minimum number of photons to give any output at all from a photosite – the other end by the number of photons which saturates. It follows that any given sensor will therefore only have one, unique ISO rating for such maximum dynamic range.

    Add gain (say 6dB), and think what happens. At the lower end you won’t gain anything in extra response, though the lowest step will become brighter on the output. But you won’t reveal a step below it – such a step would correspond to only half the no of photons required to give any output. It will remain black. At the other end, the gain will force the top step “off the scale” – it will become indistinguishable from the one below it. Adding gain in s-log mode will therefore have the effect of lowering the overall dynamic range by a stop for every 6dB.

    Andrew’s third point – that 8 bit UHD (actually QFHD) will downscale to give 10 bit is maybe more interesting. His argument is initially strong – but he is forgetting something. His mathematics assume that the 8 bit values of the four (QFHD) pixels which will go to yield an HD pixel vary randomly around the “true” value which 10 bit rendition would give. (So if the “true” value was 5, and four 8 bit values gave 4,4,4,and 8, averaging the four would indeed give 5.)

    That may be the case without any compression. But compression could easily destroy that randomness, simply through the DCT rounding within blocks. So in the example before, *pre-compression* you may well have values of 4,4,4,8 for these pixels – a compression algorithm based on DCT may well round the block values to 4,4,4,4! In which case the whole QFHD 8 bit to HD 10 bit theory goes out the window! (*PRACTICALLY* shooting QFHD to get good HD may well be a good idea. But the theory is a lot more complex than Andrews simple statement: “[I]8bit 4K down-converts to 4:4:4 with 10bit luma anyway, it’s proven by the maths[/I]”)

    So whilst I am far more convinced by Alister’s grasp of theory than what Andrew puts in his response, let’s get back to my first comment – where I soundly *agree* with Andrew. At heart is the question just how bad is this effect on the FS5? We’re talking about a high ISO, so surely Alister’s right to say you can’t expect perfection, no camera will give it, which echoes my point earlier of what such as an EX1 would look like at ISO 3200? To answer that, then just imagine a noise late at night, when you’re trying to get to sleep. Which would annoy you more – a given level of traffic rumble from a distant motorway, or a MUCH QUIETER sound of a baby crying?

    And that may be the analogy here. If this is caused by temporal noise reduction, it may well be the case that the artefacting being caused is in some cases more *objectionable* (whilst being less overall) than the video noise it’s suppressing. At least in some modes.

    We’ve been here before, with the “noise ghosts” of Panasonic’s HPX370, as mentioned before. (Though in that case the problem was worse, being at low ISO values as well.) Eventually, it was resolved by Panasonic acknowledging the issue and making the suppression switchable. It was menu selectable whether to forsake the extra sensitivity the 370 had promised – and lose the “ghosts” – or risk them for lower noise figures. I suspect we may see a similar outcome here.

    So following on from that, then whilst I query much of what Andrew says regarding theory, the statement I wholeheartedly agree with is “Simple suggestion – how about that OFF switch for the noise reduction!”

  8. Jiri Vrozina says:

    So many “Internet Experts” tested camera before it went for sale.
    Where are they now??

    HDW : The camera is fine the “Internet” footage tells its own story and I have not herd of anyone taking their FS5 back !

  9. Jiri Vrozina says:

    My friend has returned it Today.

    HDW : Now there’s a surprise !

  10. Chris Young says:

    Spent half a day with the particular FS5 Jiri mentioned that was being returned. The disappointment is that the appeal of using FS5 UHD footage on an HD timeline is gone, at least with the FS5. Punching in to a 1920 HD frame for a close up etc makes the problem even worse at any ISO over 3200. We are seeing the problem only in UHD mind you. Interestingly punch in to the 60Mbps XAVC-L footage off a UHD upgraded PXW-X70 and the issue is not there at say 18/21dB. So it’s not a codec issue by the looks of it as the artefacts are present on the FS5’s UHD HDMI output.

    Something to do with the marriage of this robust four year old S35 sensor and the new processing circuits and noise reduction algorithms some are suggesting.

    That issue aside there is another problem. That’s when you are using the 18-105 kit lens. 90% of the time if you zoom in, manually focus and then zoom back out quickly to re-frame your shot the moment you stop the zoom back the 18-105 jumps out of focus. We have tried different 18-105s and always the same problem on the FS5. Strangely if you just tap and I mean just tap the zoom servo rocker on the handle the lenses will jump back into focus. This has been demonstrated to Sony on one of their own demo FS5s. This behaviour is not present if you use the zoom servo on the side of the lens.

    Put the 18-105s on an FS700, FS7 and you have do not have that particular loss of focus issue when using the camera’s zoom servos.

    Put on the 28-135mm FE full frame zoom lens or the original old black servo 18-200mm E-Mount lens on the FS5 and the problem is not there. Sony claims that the 18-105 is parfocal and on the A7S, the FS700 and the FS7 the lens does indeed hold focus as you zoom back. It is parfocal when manually focused at its maximum reach. How Sony can specify the 18-105 servo zoom lens to be the “kit” lens is rather surprising in light of its current behaviour on the FS5.

    It will be interesting to see how quickly these issues get attended to by the big ‘S.’

    Chris Young

  11. Jiri Vrozina says:

    I wonder if it is fixable by software alone.
    It is also safe to say it is better to spend extra $$ on FS7 if 4K is required….or another brand of camera for more money.
    Either way,it is a lesson for everyone to test cameras themselves before they buy.
    OR
    enjoy tests done independent experts who just do “test” single brand of camera. ?
    By expert i also mean real cameramen who actually studied cinematography and also make living as a cameraman.
    I am always suspicious reading reviews by experts who test only one brand of camera. They never test or try anything else. All their lives. What a boring World they live in. It is like eating Lobster every day, till they die.

    Enough of that, had my say. It must be tough times for FS5 owners and now will see how mighty Sony China will deal with this.

Post Comment

Please note: all comments are moderated by an Admin.


%d bloggers like this: