Sumix 2/3" 1920x1080 CMOS

[Jose A. Garcia]

We have to wait for the new software to have live filtered preview while recording to raw.

You're right. So far the camera doesn't handle highlights very well. Let's see what happens when the new software is ready and we can do the automatic color balance via Macbeth chart.

[Paul Curtis]

Originally Posted by Jose A. Garcia

We have to wait for the new software to have live filtered preview while recording to raw.

You're right. So far the camera doesn't handle highlights very well. Let's see what happens when the new software is ready and we can do the automatic color balance via Macbeth chart.

Im not sure about the camera not handling highlights very well, i suspect is more down to being able to get the right exposure in the field, more a software thing. The RAW sensor has a good range but without knowing where your levels are - after a LUT - it's quite difficult to ballpark.

Light is log, the sensor is linear and all the colour detail is low down and needs to be corrected for.

cheers
paul

[Daniel Lipats]

Problems with highlights with this camera not like anything I had encountered before. They seem to appear everywhere and difficult to control. Whatever the cause, images from this camera do demonstrate a problem with highlights that I hope will eventually be addressed. Such problems do not seem to be present in the SI-2K for example. As expected, on extreme highlights little to no information is retained.

The image I posted is a bad example, it was not properly composed, white balanced, or lit. Thats why I hesitated to post it earlier, but I figured its better then nothing. It was done in a hurry, the only purpose of it was to test the prototype camera concepts in a production environment.

I agree, the best way to record data from the camera is to preserve Bayer. Debayer in post with more complex algorithms, and only have a simple real time preview. Not only will this allow for higher quality images but also relieve some stress off the CPU during image acquisition. In addition it would add powerful features to the post workflow.

I'm sure machine vision lenses are sharp. But I'm concerned with other things like CA, color, and distortions especially while pulling focus. I may be wrong, but I'm not sure if machine vision lenses are optimized to our needs. I would be interested in seeing more examples. I think they have potential for sure, as most 16mm lenses found on ebay will be older and certainly there has been a lot of progress made in lens manufacturing and quality control.

[Paul Curtis]

Daniel,

I'll have to see if i can dig some sample images out from my tests.

when you preview what you are about to shot, what are you actually looking at? Raw sensor information (very dark) or a display that has gone via a LUT?

I understand your test image, but real world unprepared examples are often the best!

If you're going via a LUT then perhaps that LUT needs work. I found that you expose for the highlights and be mindful of what's there in the shadows which is where your real midtones lay. But unless you're looking at full sensor data who knows what has happened to the image?

I have a stack of tests and things that i did but no time to write them up as i intended. I will see if i can find a couple of images that might represent what i mean.

See if you can borrow one of the fujinons, it really was pretty good (better than the 16mm cookes and switars i was also playing with). Nice feel to it too. Of course you may need to ensure that the lens you get is for single sensor and not 3 CCD (which would probably give CA)

cheers
paul

[Paul Curtis]

I've dug out a sample of what i mean (my first attempt at uploading images as well...)

the uncorrected image is basically linear result from the sensor and all the information we're interested in is in the shadows. These were 16bit RAW bayered images, debayered (badly!) into 16bit RGB tiffs and then in AE use the cineon Linear to Log for a quick and dirty conversion, some CC tweaks and then output back to 8bit.

Highlights are protected (see the window, top middle) yet there is good range in the shadows, the colours are pretty accurate for the day. The colour sparkling around the branches is just a result of the simple bilinear debayer. They were originally 1080p but i resized them for here.

Lens was a fujinon 25mm f1.4 however that's nor fair because it was a 2/3rds lens and the image circle barely covered this sensor (which was 1"), the edge performance is soft but that's the reason.

I hope the images look okay, my main workstation is calibrated but this laptop isn't and they look too wishy washy on here...

hope this helps...
cheers
paul

[Daniel Lipats]

With the software that comes with the camera you see exactly what the image will look like once captured as a still or a video. However, there is no real time preview while capturing. You can use the histogram to help expose.

For the image I took that day, I only exposed by what I saw on the lcd. If I saw highlights I would step down and try to find a balance. This balance ended up at a point of strong highlights an too dark shadows. data lost on both ends, but I did not have a chance to do anything more. Had to be quick.

Except for the soft edges and a bit of a hot spot your images look good.

[Thomas Richter]

Paul,

let me see if I got you right. Depending on many factors (monitor gamma, LUT, filming environment) we may see a scene that LOOKS underexposed. Yet, it is correctly exposed as we are shooting to protect the highlights.

We then still have enough latitude and low noise levels, so we can pull the mid-tones in post.

If that's what you are saying, it would be great if Jose could repeat the balcony test (with the flowers). I would love to compare latitude to the SLR again.

Please keep posting,

Thomas

[Paul Curtis]

Originally Posted by Thomas Richter

Paul,

let me see if I got you right. Depending on many factors (monitor gamma, LUT, filming environment) we may see a scene that LOOKS underexposed. Yet, it is correctly exposed as we are shooting to protect the highlights.

We then still have enough latitude and low noise levels, so we can pull the mid-tones in post.

If that's what you are saying, it would be great if Jose could repeat the balcony test (with the flowers). I would love to compare latitude to the SLR again.

Please keep posting,

Thomas

Thomas, Im just attempting to show the kind of image that comes directly off the sensor. Im concious that this is a summix thread and don't want to pollute it too much. Also light works logarithmically, so a sky is many orders of magnitutude brighter than a skin tone and being able to capture both is fundemental to all cameras. Even in your consumer cams the sensor is probably pulling 12bit off the sensor it's just that the AD converters and compression result in 8 bit HDV files simply because it's easier to transport.

But the dark image is closer to what the sensor sees and once you get RAW from the sensor that could be what it looks like. However in summix case they are sensibly applying a LUT or Gamma adjustment so that the dark image is processed a bit and you can see into the shadows. But a LUT should be more than a gamma curve and you'd want to change it for different situations. It would be nice though to be able to flag or mark certain exposure areas on the display - i've always fancied a zone system with false colours overlaid.

However one of the appealing things about these DIY cameras is that you do have control over everything so i would personally want control over the type of lookup in the camera or have the frames RAW. I hope that with summix you can.

I hope that makes sense. You'll only be able to judge the performance of the summix once RAW output is enabled, but it should be good. However none of these camera are magical!

cheers
paul

[Thomas Richter]

Paul,

I am fully supporting your argument, just trying to summarise it. It is the linear sensor response that we have to cram into a log. Depending on how steep the log curve is, midtones may still appear dark.

I understood that the poor highlight handling may just be using the wrong exposure settings rather than a shortfall of the sensor - just wanted to reconfirm that understanding.

On a side note, isn't it interesting how our body copes with variation? Log pattern hearing response, log pattern brightness response ...

Thomas

[Take Vos]

We probably even have a logarithmic pain receptors. I mean you can feel the lightest touch when only a single hair on your arm is moving or drive a thumbtack into your arm, and anything in between.

[Régine Weinberg]

and what will happen then?
if it will be reality
i do not have any doubt about
http://redmodz.com/component/content.../84-3k-scarlet

have fun

[Thomas Richter]

Originally Posted by Régine Weinberg

i do not have any doubt about
http://redmodz.com/component/content.../84-3k-scarlet

I would not be trying to contribute to this thread, if Scarlet had featured an exchangable lens. But its great, isn't it. Sumix has a chance to get a large piece of the pie now.

And what's the fun of having a cam everybody else has ;-)

[Jose A. Garcia]

You're right Thomas... Scarlet sounds almost too good to be true. On the fixed lens part, I'm afraid Red won't put a cheap piece of glass there, so that's not really a big problem. I was wondering, is it really so important to change lenses when you have a high quality 2/3" specific 2.8 8x zoom? Ok, a bit slow, but as you say that's the strong point for the Sumix appart from the feeling that you're using a camera that very few people is using for filming purposes (that's encouraging for us DIYers), but then at less than $3000 for the basic Scarlet we better be ready to see birthday kids shot in 3K digital cinema.

EDIT:

I was thinking of this last part. Red is actually offering the camera for what they call "Soccer moms". But when it comes to "take the footage to Windows Movie Maker (lol) and add your kid's favourite music", how's a standard family computer going to edit 3K? I supose the cam will have 1080p mode but then the soccer mom won't have wide angles.

[Daniel Lipats]

We have yet to see any footage from Scarlet. It may not be all that impressive. It has some very small pixels which in theory will have an impact on many factors including sensitivity.

I'm not trying to discourage anybody. I myself will probably buy one when its available. Really looking forward to seeing some footage.

I think its way too early to throw in the towel. We should be able to get a higher quality picture through the Sumix camera. Especially by having much less compression.

[Jose A. Garcia]

I still have faith in the Sumix. And even more now that I know it'll become a pc-independent cam.