Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Got some artifacts probem low framerate stills, is this normal for dragon?

Björn Benckert

Björn Benckert

Well-known member
Joined
Dec 5, 2007
Messages
7,423
Reaction score
74
Points
48
Age
52
Website
www.syndicate.se
We are shooting some phone company product films every other months and we want to shot stills with the same lights as we do for motion. And I try to push as much out of the dragon sensor as possible for those stills but I run into a problem I seen before. Basically it´s long exposure and really high Fstop and super low iso.

Then when looking at the result there is "broken pixels" mostly pure blue or red, but also yellow and other colors and it seam to happen along edges like on the computer sceen edge in this test frame on the link below... is that normal?


I shoot 4fps 180 deg shutter, with correct blackshade, done it at the right temperature etc. So no it´s not a blackshade thing. And I shoot frame burst so no compression. And I shot focus brackets with focus increments of 10cm at the time and it seams like the error is some what constant between the takes and also between different takes.

Here is a we transfer link to a highress tif, it´s exported with ADD and the artifacts are looking just like it in the r3d´s. And I seen this behavior on two differnet dragons. Last time I shot 1fps 360 shutter so I figure, well thats a bit of the edge of what the camera can do and I though that was the issue, but 4fps 180 deg is not really super slow and the amount of artifacts seems a bit high I think, no?

http://we.tl/qdvMISdLEP


When downscaling most of it goes away but these guys think 6k x 8k is a bit small so I dont want to deliver 4k or such. With a bit of luck I get time to do the same thing again and roll at 8fps and a stop less to see if that helps but seems strange that the camera get these problems, no?
 
Any dslr frames to test against. Just a Nikon or 5D, just for reference.
 
David Battistella said:
Any dslr frames to test against. Just a Nikon or 5D, just for reference.
Click to expand...

Nope,
They might behave the same but I dont think so 1/20 of a second on a SLR usually does not create any issues. I seen similar stuff on SLR's when doing seconds of exposure but not really when shooting fractions.

But when you say it I will bring a Mid format camera to shoot next to it just to see the difference in sharpness etc.
This we shot with the sharpest 35mm lens I had my hands on, a really nice PL rehoused Leitz 35mm macro.

But I really wish I did not have to retouch away all those dead pixels. And we need to be stopped down all the way, f16 is not really enough so we shoot focus passes for the screen and front of the keyboard. But I guess I can open up a stop and shoot 3 focus passes at double the speed and the dead pixels would be cut in half or hopfully go away completly. But it seams odd that the camera can not do clean frames at 4fps, I dont remember our dragon having this problem but these two cameras that I used since our got stolen both have these dead pixels.

If it was me I would not care much just downsample and then its gone... but you know clients they will for sure zoom in there 1:1 and point at every one of them :)

Here is a link to a closeup of the problem, as I realize the big file is a bit big to download. Basically those dots on the macbook logo sensor issues.

https://www.facebook.com/photo.php?...523042.-2207520000.1440714317.&type=3&theater
 
Yes, I have seen these way back when.

A DSLR frame wouldn't solve your problem.

I'd shoot Graeme an r3d.

You have compared to a non ADD frame, I guess?
 
Last edited:
What stands out here is that the bad pixels are very tightly clustered around strong edge boundaries in the scene. It could be that an algorithm that cleans up aberrant sensor pixels by relying on their neighbors becomes useless at the edge transitions. In those areas, the jump in scene brightness becomes indistinguishable from the jump going into a stuck pixel. The algorithm then has to fall back on letting it all through so as to preserve the edge. The logic probably already contains a refinement to look at the individual channels. If this is the case, there may be not be a lot Graeme can do besides some refinement. Otherwise, the process would have to jump to a content-aware guess like Photoshop's, with mixed results.
 
Hm... When I have seen it, it hasn't been nearest-pixel cleanups, though that could sound like an ADD thingy.

That could probably be tested through running a non-add pass.

It has also not been stuck to the pixel-matrix, but rather non-uniform between frames when I have seen this.
 
Last edited:
Gunleik Groven said:
Hm... When I have seen it, it hasn't been nearest-pixel cleanups, though that could sound like an ADD thingy.

That could probably be tested through running a non-add pass.

It has also not been stuck to the pixel-matrix, but rather non-uniform between frames.
Click to expand...

its in the r3d´s with or without add.

I can upload a r3d later today but it looks just the same.

And yes there is something special going on around edges between bright and dark but then there is also pixels on flats that just sticks out in signal colors. The linked tiff we retouched some of it all ready but if you zoom in 1:1 and pan around you see there is still a lot there.

I know most sensors has bad pixels and get a factory setting to steal from it´s nearest neighbor for the bad eggs. This kind of looks like a lot of those and like such fix is out of place or missing the target or such. But I think it´s has something to do with the exposure time as I doubt it would be there if it was shot 25fps something i would need to test when I get the time.
 
I just had a look.

The pattern looks like it has some order to it, like every X number of pixels a R G or B isn't responding correctly or turning on or something.

This tends to show up when sharpening is applied, but an r3d would be interesting to look at. I've seen this as well, but very rarely. Definitely get an r3d to RED. You might have to do some pixel busting on that frame though.

David
 
David Battistella said:
I just had a look.

The pattern looks like it has some order to it, like every X number of pixels a R G or B isn't responding correctly or turning on or something.

This tends to show up when sharpening is applied, but an r3d would be interesting to look at. I've seen this as well, but very rarely. Definitely get an r3d to RED. You might have to do some pixel busting on that frame though.

David
Click to expand...


No it´s there before sharpening the sharpening enhance off course but they are very much there before it´s applied.

here is a r3d link: http://we.tl/0jTBimO1Zc
 
Interesting. I saw these on the proto I had back in 2013, but the situation that triggered them was nothing like this, it was in high frequency situations with plenty of detail.

Definitely file a ticket with RED on this.
 
Now here is a new frame with half the exposure time i.e 8fps 180deg shutter and a blackshade to go with it. Then the problem is far less apparent, but some pixels are still out of line, so I´m pretty sure it´s exposure time related.

http://we.tl/pQXioRTGZd

We shoot a lot in really low fps and normally this problem is not a problem as we then Downsample to HD and all those dots are then washed away in the interpolation of the downscale. But when staying in 6k it´s really something to look out for.
 
Is this for still photography? This is usually why I don't buy the DSCM concept. Yeah, you can do photography with the Dragon, but there are better options for still photography. If I were to shoot product shots I would use a medium format digital still camera with flashes, I always get better result with that. But keeping with the same lights I would still use a proper still camera.

But if it's for "stop motion" or your motion control rig shot at one frame at a time it's a not very good. Maybe the 8K will be better off for these kinds of gigs, downsampling to 7 or 6K.
 
Christoffer Glans said:
Is this for still photography? This is usually why I don't buy the DSCM concept. Yeah, you can do photography with the Dragon, but there are better options for still photography. If I were to shoot product shots I would use a medium format digital still camera with flashes, I always get better result with that. But keeping with the same lights I would still use a proper still camera.

But if it's for "stop motion" or your motion control rig shot at one frame at a time it's a not very good. Maybe the 8K will be better off for these kinds of gigs, downsampling to 7 or 6K.
Click to expand...

This is for moco and stills and the whole idea of having the exact same colors, look, light etc for the highres stills and motion stuff is working great. and the quality is, if done right on par, or for some stuff even better with the dragon than the mid format cameras, atleast for what we do.

But for this case we want to shoot with as high Fstop as possible to get a deep enough macro DOF it demands quite a bit of light... like enough to blow a fuse in the studio and turn the place into a sauna.

Then long exposure is a nice thing, as it brings the amount of lights down. And for moco the slower you move the better you are. So if this was no issue then a lot of this stuff I would shoot at 1fps 360 shutter with x100 length of the wanted clip. As doing so makes lightning easy and the material does not really have a frame rate per say, as then we can compound and get out any motion blur or frame rate.

So yes for us low framerate is key. And it works fine for HD/2k all these artifacts are then interpolated away... but then the highress stills needs to be retuouched which is not ideal.
 
I wonder what would happen if there was an actual image on the screens? What I am getting at is that maybe with the flat fields above and below (bands of color, black, grey, pink), it might have triggered a weird compression artifacting. Phil's rainbow is a good example, but there are subtle gradients in the "transition" between colors whereas this is a very hard transition.

X number of pixels should just be black.

I would file a ticket with a log file as there might be something going on with one of the boards or something, Then I would do a complete system restore and a new BScalibration, just to see if it reproduces the problem and I would also do a series of compression. 2:1, 3:1, 4:1, etc to see if the artefacts are happening on the same pixels. If they are then it might be a sensor thing.

definitely there on the R3D.

whenever I capture purely for stills I tend to use from averaging and burst mode, this gets you cleaner results than just frame grabbing from a motion stream.

I don't know how many frames in your moco move but why not set it as a time-lapse with 2 frame averaging and snap a shot every 3 or 4 seconds this way you could probable get whatever Fstop you want with a 1/50 shutter and still the security of every frame being captured as a still frame.





David
 
For long exposures such as this, if the camera is locked off it's best to keep the exposure length "normal", but engage high fps, frame averaging (up to 16 frames for best results) and let the ISO go up as high as you need to get the shot (even up to 6400 can look superb). http://www.graemenattress.com/wp/160-pg-unicorn-2/ were taken via this method (I'm often stopping down to get the DOF I want) mostly at ISO3200 and ISO6400, but sometimes up to ISO12800.

Graeme
 
Just so we know, what would be a good between thumb and index finger "lowest exposure" to aim for if we don't want to risk broken pixels. ?
 
this one was 1/1oth of a second at 144degrees. The only way is to test.
 
David Battistella said:
this one was 1/1oth of a second at 144degrees. The only way is to test.
Click to expand...

But the issues remain at 8fps 180 deg.. as on this one, even though it´s less.

http://we.tl/pQXioRTGZd

As we want to shoot f22 preferably and be able to use it for film and stills frame summing is a bit of a mess so would be good to know where.
 
You must log in or register to reply here.
Back
Top