Daniel, Does the codec software confuse the order of the colors? We do not see this problem in our Bayar data and deBayer images. Perhaps the codec is assuming a different order of GRGB than the sensor is putting out.

Farhad,

When selecting the CineForm RGB Codec, you have to specify the pixel order. I have tried all 4 pixel order settings, this one is the only one that works at all. Its also the same pixel order as in the documentation.

I don't think there is a problem in the Sumix Bayer data. The videos the Sumix software outputs seem to display correctly.

I am working with NorPix to resolve this issue.

Here is a bit of an update...

First of all, the bug with cineform has been fixed. New versions of cineform should work fine now. I have not had a chance to test it yet.

I did have a chance to test some new software from Sumix for the camera tonight. Its getting better, has more features and becoming easier to use. Has a new RAW capture mode, along with the old AVI. I was very happy to see zoom options, so now the video can be scaled down to fit on lower resolution monitors without cropping.

However, it still lacks a few basic features which are very important to our needs. Here they are summarized:

1) Preview
When recording, the video preview shuts down. No way to watch what your capturing.

2) Recording
You still have to specify how many frames you will record ahead of time, for example the default is 100. Once you hit record, the software captures 100 frames and when complete it dumps that to the hard drive. I can see the need for this because a single hard drive is simply too slow. A raid setup would be required for real time capture to disk.

Right now the Sumix software supports an uncompressed AVI format. If a 3 second uncompressed AVI video is ~600mb, then 60 min of video is about ~720gb.

I think we have to consider some form of compression.

but still we should have the possibility to shot uncompressed, an option to choose is always wellcome

Paul, I would imagine this has to do with the fact that most of the CMOS camers are single sensor design. So they use a Bayer sensor pattern that over-represents the greens at the expense of red and blue. Hence raw images will look greenish and desaturated.

On the other hand, most of the CCD cameras are three sensor design, so you are getting the full amount of color raw.

This may be the explanation for what you are seeing.

Not quite true . . . just look at the Viper, a 3CCD camera that still has green as the most sensitive color. Also the Andromeda modification for the DVX100 showed magenta highlights where you find that the green has clipped before the red and blue channels.

The "green" can be very simply removed by white-balancing though. So when you see a "green-ish" RAW image, it simply means there is no white-balance applied.

Thanks,

Jason

I have been playing with different settings and im starting to finally get some good, clean images from the sumix camera. Its giving my Panasonic DV camera a run for its money in terms of dynamic range and sensitivity.

Here are some tests I shot this morning. Set at 12bit, 24fps, 0 gain. Unfortunately I forgot to white balance! So they all have a bit of a green tint to them. I tried to remove with some post work.
http://dreamstonestudios.com/delete/frame63_zeiss.jpg
http://dreamstonestudios.com/delete/frame_lens2.jpg

Here is a still from the same setup taken with a Panasonic 3CCD DV camera:
http://dreamstonestudios.com/delete/IMGA0245.JPG
http://dreamstonestudios.com/delete/IMGA0241.JPG

I think my next move should be getting my hands on a real HD c-mount lens. The images are a bit soft, and I think they can be improved on a lot.

I'm starting to like the results.

Good to see you are still working on this. The images look nice but i cant tell if the edge blurring is from poor lenses or odd bokeh, and is it just me or does it look like there is still some infrared light being picked up? i remember finding that without correct IR filtration on my old sumix camera images had a similar reddish tint and a some diffusion from the IR light not focusing correctly. it doesnt look that bad on yours compared to my old camera but it does still appear to be there at least to my eyes. im not even sure how you can compare to your panasonic (gs series?) dv camera... :P

Noah,

I think its just poor lenses. Something odd about this Zeiss one especially. The picture with lens2 prefix is sharper. I'm looking to spend $300-$500 on a decent c-mount prime but I can't find anything I like. I don't want to make another uneducated purchase, I have spent over a grand on poor lenses already.

The reddish tint is actually my rushed color correction attempt. It hurt the color quality a bit but does look better, it was pretty green. I will be shooting more tests and will make sure to white balance.

Yeah, I think its a GS150. I wanted to show what the scene looks like through a CCD camera and its the only one I have here at home. I hope to put it head to head with prosumer HD cameras soon, but first need better optics and a faster network card.

Daniel, how is the camera performing with low light levels? when does grain appear?

I am getting ready to do a new set of tests this weekend. Hopefully I will have time. This time I will attempt to get a better idea of the dynamic range and will take it through a variety of gain and light levels.

However, I don't believe that the full potential of this camera has been reached... So, i'm not so sure that any of it should really hold any weight. I am now considering shooting a film with it, so tests should give me an idea of how far I can push it in its current state.

I got a C-Mount -> Canon FD SLR mount adapter for ~$30 and took a few min to try it out. I did not have any Canon lenses at home, only Minolta so I hot glued them onto the end.

http://www.dreamstonestudios.com/delete/boat1.jpg
http://www.dreamstonestudios.com/delete/boat2.jpg

50mm SLR f/1.7 Minolta set at f/4... 0 gain, 12bit, 24 fps, I think it was at 36mhz. Behaves more like an 80-100mm on this camera.

Looking at a chart here...

35mm -> 16mm
-------------------
14mm | 5.6mm
16mm | 6.4mm
20mm | 8mm
24mm | 9.7mm
28mm | 11.3mm
32mm | 13mm
40mm | 16mm
50mm | 20mm
85mm | 34mm
100mm | 40mm
135mm | 54mm


I took some Canon lenses home, a 50mm and a 28mm. I would like to find a manual Canon FD 24mm or even a 20mm.

36mhz is the shutter? how was it lit? did you had any chance to test the latitude? when does grain start to appear?

No, had no time to do anything other than throw the SLR lens on and see what that looked like. We have some clients right now keeping us pretty busy. Should have more time after this week.

36mhz is the shutter. The image is a bit brighter with 36mhz but there is a little rolling shutter distortion. IMHO its in tolerance, others may say otherwise. It would depend a lot on the situation. What your shooting, how its moving, how the camera is moving, and what light you have to work with... But I would not go any lower.

It was lit with ambient sunlight coming through windows and a 1,000w softbox about ~5 feet away.

36th of a second is 36mhz?

36Hz would be 1/36th sec... 36mhz is 1/36000000th sec :P . are you talking about shutter or the pixel clock? my guess would be that you mean you are running at 36mhz pixel clock, and altasens does 2 pixels per clock... so 72MPps which at 1080p and 8b works out to ~34fps? this would probably give you a default shutter of almost 1/34th of a second since machine vision type cameras tend to default to full open shutter for best exposure...

total shot in the dark, but is that what you mean? :)

Yeah its the pixel clock frequency. ~34 fps sounds about right. You can program it to expose at 24 fps, which improves light sensitivity.

Light sensitivity at both 50mhz and 36mhz is the same as long as the framerate is the same.

Hello everyone,

I've been following this project for some time but never posted on this before. It seems this thread is quiet for now. Is Sumix still developing the software toward digital cinema uses? What's the status of the current version?

It isn't that Eric, it is just that Sumix, despite following the steps of SI, doesn't have a clue about how to put the minimum required features on the firmware/hardware for being able to develop a successful approach.
Maybe they are just gathering experience to move to a higher goal...

From what I've gathered the camera head is capable of dumping 12-bit RAW data, and recording to Cineform RAW format is already working using StreamPix. So I would imagine only a more photographer-friendly software/UI is needed to do some basic shooting. Or am I missing something?

I have some coding background but nothing related to machine vision cameras. If I were to do it I think the easiest way is to record uncompressed RAW. The data rate of uncompressed 1920x1080 24p 12-bit RAW is around 600Mb/s, about the capability of a 3-HDD RAID 0, not so unthinkable nowadays. Cineform is great but to me the added cost is considerably high.

Of course there are hundreds of other features need to be added, but if things are on the right track, I think those can all be done in time. The idea of a SI-2K mini for 1/5 the price is very tempting. I hope this project will go on. I'm seriously thinking buying one when my bank account allow.

what do you guys think of "Panchromatic" cells? couldn't this be applyed to the summix camera? Bayer filter - Wikipedia, the free encyclopedia

there is a reference to an early version: Image:RGBW Bayer.svg - Wikipedia, the free encyclopedia wich i think it's the best of those, because there is equality in all color photosites plus a white photosite

this doens't give 4:4:4 but still al colors have the same amount of photosites plus an added bonus of more light because of the white one, wouldn't this be easy to add to the camera?

this wouldn¡'t give you more resolution but more light sensitivity

Hello Daniel,

This may be a little too late, but I think the reason your Cineform RAW AVIs look different from SI-2K's ones is because SI applies a 12-bit linear to 10-bit log transfer before the encoding while yours are untouched. SI does it to preserve the most of the 12-bit linear data into 10-bit, and though the image may look right, it isn't. Not mathematically anyway. It just so happens that the linear2log curve is very similar to a gamma curve.

A true linear (gamma=1.0) image will look very dark on a standard gamma 2.2 display. Though it may appear featureless at the dark area, the detail is still there. All you have to do is apply a 2.2 gamma correction in post to get a normal view. Make sure you use 16-bit/channel project setting if you are using AE, for if you clip it to 8-bit/channel you may get color banding.

Here are some test images of your clips. No CC was applied other than the 2.2 gamma correction.

Eric,

Yeah you are correct it is the gamma. There is a Gamma Correction option in the new software which appears to have the same result. The last few images I posted is with the option enabled.

Sumix is currently working on a software update with more features and better designed for our application.

what is the strange colors in the sails (the ones that aren't in shadow) in that picture:
http://www.dreamstonestudios.com/delete/boat1.jpg
it is due to the debayering? or some effect of the softness of the lens? is it the loss of optical resolution due to the use of an adapter for lenses much greater that the ones needed?

asuming that the effect is due to the lens, if some lens can make those kinds of effects in hight frequency areas (because the background doesn't display such effects) how can on know what lenses don't make those effects?

Well there are two things to keep in mind about the pictures...

First... I'm mixing daylight (~5600k) and tungsten (3200k) lighting. So colors are going to look a little funny. Especially when mixed in the (white) sails.

Second... The color filter in the camera is incorrect. It blocks some colors more than others and leaves a tinted and desaturated image.

I have been referred to the correct filter, but I'm going to wait for Sumix to provide the official hardware fix.

Were you referred to a third-party color filter? and is the fix provided by Sumix going to be a new filter or a color matrix (firmware) update as well?

The IRC-30 filter here:
IR cut-off filter

The filter I'm using right now cuts out red light, and when compensating with white balance it increases the noise by boosting the signal. Furthermore, the filter cuts out an additional 50% of all visible light. The camera is not performing as well as it should be.

I'm not sure which filter I have. My results are similar to:
http://www.dreamstonestudios.com/per...industrial.JPG

Here is a chart with the IRC-21 (cheaper filter)
http://www.dreamstonestudios.com/per...umix/IRC21.JPG

Finally, the IRC-30 (recommended filter)
http://www.dreamstonestudios.com/per...umix/IRC30.JPG

Images were provided by Sumix

That first color chart is TERRIBLE. Hardly any color information.

The IRC-30 gets you about about 70% there. If you add color matrix calibration, you can get a lot closer, but you can still see that the colors aren't pure.

See attached image.

Hello Daniel,

Can you confirm that the camera can be set to run 1920x1080, 24fps, 1/48 shutter speed (exposure time = 20.833ms) and 12-bit output, at the maximum pixel rate (74.76 MHz)?

I think this is the most common mode if we are using this camera for digital cinema works. To minimize the rolling shutter effect it's crucial to run it at the maximum pixel rate.

Thanks!

Eric,

I just tried setting the camera to that configuration, seems to work fine. The screenshot below will help answer your question. As far as I know, yes it is capable of running at the settings you mentioned.

That's great news, Daniel.

I noticed the exposure showed 0.769128ms, which is about 1/1300 shutter speed. I suppose it's the minimum value, isn't it? Otherwise it wouldn't make sense.

I'm planning to order one of the camera and write my own software. What I have in mind is recording uncompressed 12-bit RAW, real-time preview while recording, and a touch-screen UI like the SiliconDVR's.

Thank you for sharing the informations and doing the tests.

Eric,

I'm not sure how to interpret the exposure value. That is in fact the maximum setting. I did not try to override with a manual input but I don't think it will work. The maximum exposure value depends on the set frequency, at 10 MHZ it peaks at only 4.98 ms or so.

I'm going to try and order the IRC-30 filter tomorrow and figure out a way to mount it. The filter should make a considerable improvement to the picture quality.

I'm no expert but I think an optical low-pass filter is needed to eliminate aliasing. So maybe it's these OLPFs incorporated with IR cut-off function that you'll need:

Optical low-pass filter

is there a need of low pass filter? how is aliasing showing? any clear example? oh, and Daniel, can you post a screenshot of every segment of that "settings" window you posted above?

Aliasing and Moire patterns
Moiré pattern - Wikipedia, the free encyclopedia

I think most hi-end digital cameras have OLPF. It's likely the original Sumix filter is one, too.

yeah i know that, that's why i suggested the use of that "new" patern in the bayer filter, but i guess that is not up to sumix but up to the maker of their imagers, that patern could get the same information from half the numbers of photosites, so we would be getting double amount of information, so a pixel binning thechnique should do to combat aliassing, but i gues sumix don't print the bayern patern in the imagers of their cameras

Quick update, I ordered the IRC30 filter earlier this week and it will be coming in the mail tomorrow.

I will try to find time this weekend to shoot a few more tests.

Here are the results from earlier today, shot with the new filter. Natural light only.

http://dreamstonestudios.com/persona...es/IRC30/1.jpg
http://dreamstonestudios.com/persona...es/IRC30/2.jpg

Daniel, do you have an idea where the pattern on the corner of the clock comes from?

Hi Daniel,

These pictures are gorgeous! To my eyes the resolution and dynamic range are very good. Even better if you can upload some footages.

The colors are not very different from your last pictures, though. Which is not a bad thing IMO. Raw data directly from the sensor are not optimized for human vision. Those Sony, Panasonic video cameras do a lot of processing like color matrix, white balancing, curves, before outputting the image. But those can all be done in post. recording raw data keeps the maximum flexibility for post processing.

There seem to be debayering artifacts. I'm not sure if that's what Gottfried Hofmann refers to but I think it can be easily fixed by choosing a more sophisticated debayer algorithm.