Philip Bloom's video review of the 1DC

[Kolor-Pikker]

Great tip by Phillip never to shoot below 400ISO in log mode, to avoid banding! Thanks!

Also make sure to have lens shading disabled in C-Log, the flashing Log icon was a warning that he had it on, and it may increase noise levels. Actually, make sure to have any and all in-camera corrections or enhancements turned off.

^ The typical stills APS-C sensor can only take 3 or 5 frames per second. Make it 8-mpix and you may get to 8 fps. If you want 24 fps, it has to be a different design, one that can run faster. It will be more expensive, because there are design tradeoffs, and because it will be a low-volume part.

Stills cameras have been able to shoot as fast as 12fps way back in the film days, so it's not a mechanical limitation, not that the shutter is used for video recording.

It's not a sensor limitation either, as all modern CMOS sensors output at a fairly fast refresh rate, the 5D2 had a 30hz refresh which is why it never got 60p.

So the only logical conclusion is that it must be image processing, if the 1Dx can only now fire off as many fps as a 20 year old film camera, it must be that image processing must have improved enough to allow it to do so.

The 1Dc is unique in that it's the first DSLR with a heatsink, normally the sensor isn't under such load that cooling is required, but video cameras do need some cooling.

In order for a camera to put out a good image from the sensor, it must be able to debayer and scale the whole frame of pixels down to the output size, at x frames per second. That's a lot of data. Without a dedicated DSP for video, the a cine DSLR would need much more processing power than the 1Dc to output the whole sensor data, especially if you want efficient compression.

 

[Samuel H]

^ Not true, AFAIK:
* mechanical shutter is indeed a limitation, specially in full frame cameras: they've gotten much cheaper, but not really much faster than what was out there 20 years ago (http://www.canonrumors.com/forum/index.php?topic=11952.0 --> 1DX: "The shutter limits it to 12 fps")
* stills sensors are actually quite slow, the only way they get to 24-30-60 fps is by line-skipping

The datasheets for the sensors in the cameras we use are, in general, not public. But you can, for example, check this very modern Aptina APS-C sensor, which can do a maximum of 10fps: http://www.aptina.com/products/image_sensors/mt9h004/
It's a 10fps sensor, but it can record 30fps video using some trickery, in this case cropping (16:9 instead of 3:2) and binning (2x2 to 1). Other sensors don't get such a speed boost with binning and use line-skipping instead.

And regarding processing, consider the following: the Digic DV III processor that the C300 has can get a ~4K image and convert it into a ~2K debayered, subsampled and encoded image, at 30 fps. That processor can also be found in the $400 Canon Vixia HF R-200. Yes, that's how high-end and super-expensive that processor is.
I know that a processor that's designed for something else (stills) can't match the performance of a tailored solution, but really, I don't think the processing is that demanding. And I know the sensor readout speed is an issue.

 

[lumiere61]

Sorry but still the question is ... technically could you crop the sensor in the 5D mkII at same way of the 1DC or not ?
I am pretty sure that the answer is yes.
Is this crop reading more critical for the sensor compared to line skipping ?
I do not know but I would like very much to know
Now that I think about yes ,,, it could be very very critical for the sale of the multithousand dollars videocameras.
Thanks and regards,
gino

 

[Samuel H]

yes, you can usually crop the sensor
but no, it's not more efficient than line skipping: usually, readout time is a fixed number of nanoseconds times the number of rows you want to read
it seems like with the 1DC the S35 crop is enough to read all the lines you need; with the 5D2, which IIRC skips 2 lines out of every 3, the crop would be massive

 

[Cosimo Bullo]

And these clips are where? If you watch stuff that has been already compressed (youtube, vimeo, compressed h264 out from an NLE) you are looking at that compression. At the moment every out of focus background will fall apart. Even in bluray discs, you might get banding in foggy scenes, no matter if the camera was red or alexa. Because of the compression.

https://vimeo.com/58252561

The detailed stuff looks great, but the out-of-focus areas seem to be falling apart (I have downloaded the 1.5 gig version). Upper left at 00:13 seconds for example.

 

[ReelFilmGeek]

https://vimeo.com/58252561

The detailed stuff looks great, but the out-of-focus areas seem to be falling apart (I have downloaded the 1.5 gig version). Upper left at 00:13 seconds for example.

I have a feeling that shot was at less than 400 ISO where at the moment the camera does fall apart. Philip was actually talking about it at the Miami workshop and was saying how he was trying to use film grade to hide the poor bits of the footage. It has something to do with the fact that the camera has a native ISO of 400 and the way the camera simulates lower iso.

 

[lumiere61]

yes, you can usually crop the sensor
but no, it's not more efficient than line skipping: usually, readout time is a fixed number of nanoseconds times the number of rows you want to read
it seems like with the 1DC the S35 crop is enough to read all the lines you need; with the 5D2, which IIRC skips 2 lines out of every 3, the crop would be massive

It took me a little but now I think i have understood that this cropping techinique is very resources demanding
Thank you very much again for your kind and valuable explanation
I understand that this level of performance will not ever come at cheap prices. Everything must be top quality
This in unfortunate but also reasonable
Thanks and regards,
gino

 

[atarijedi]

^ Not true, AFAIK:
* mechanical shutter is indeed a limitation, specially in full frame cameras: they've gotten much cheaper, but not really much faster than what was out there 20 years ago (http://www.canonrumors.com/forum/index.php?topic=11952.0 --> 1DX: "The shutter limits it to 12 fps")
* stills sensors are actually quite slow, the only way they get to 24-30-60 fps is by line-skipping

The datasheets for the sensors in the cameras we use are, in general, not public. But you can, for example, check this very modern Aptina APS-C sensor, which can do a maximum of 10fps: http://www.aptina.com/products/image_sensors/mt9h004/
It's a 10fps sensor, but it can record 30fps video using some trickery, in this case cropping (16:9 instead of 3:2) and binning (2x2 to 1). Other sensors don't get such a speed boost with binning and use line-skipping instead.

And regarding processing, consider the following: the Digic DV III processor that the C300 has can get a ~4K image and convert it into a ~2K debayered, subsampled and encoded image, at 30 fps. That processor can also be found in the $400 Canon Vixia HF R-200. Yes, that's how high-end and super-expensive that processor is.
I know that a processor that's designed for something else (stills) can't match the performance of a tailored solution, but really, I don't think the processing is that demanding. And I know the sensor readout speed is an issue.

I don't think Aptina is a good example. Look at CMOSIS CMV12000 (12MP), it can do 150fps and is an APS-C sensor, the CMV20000 (20MP) can do 30fps, and the CHR70 (70MP) can do 3fps. There is a sensor called the Alexima AM41, it is 4/3", and can do just over 500fps. These are all in global shutter modes, the frame rates increase when you move to rolling shutter, and if you are just doing stills, the dynamic range on these sensors can go up from 11/11.5 to about 13-15. Fairchild Semiconductor developed an sCMOS sensor, similar to the one used in the BMCC, it is 5.5MP and can output 100fps, very high DR, and very, very, low noise.

The issue isn't the sensor read out, it is the processing system, but not the initial processing, as debayering and chroma sampling aren't all that processor intensive. Compression is, and especially if you want high quality, and high speed, as you get into systems that need multiple simultaneous processing engines, and it gets expensive to develop!

It'll drop though, and not that far off either. Especially with sensor companies like CMOSIS, and Fairchild pushing relatively decent image sensors at low costs, and companies developing camera solutions using FPGAs instead of custom ASICs and general purpose DSPs like TIs Keystone DSP multicore system.

 

[maarek]

https://vimeo.com/58252561

The detailed stuff looks great, but the out-of-focus areas seem to be falling apart (I have downloaded the 1.5 gig version). Upper left at 00:13 seconds for example.

That 1.5gig is still not the original file. It's wayyy compressed compared to the original stuff coming out of the cards. Though that part does look pretty horrible anyway. Maybe it was the ISO 400 thing in log-mode?

 

[Samuel H]

I don't think Aptina is a good example. Look at CMOSIS CMV12000 (12MP), it can do 150fps and is an APS-C sensor, the CMV20000 (20MP) can do 30fps, and the CHR70 (70MP) can do 3fps. There is a sensor called the Alexima AM41, it is 4/3", and can do just over 500fps. These are all in global shutter modes, the frame rates increase when you move to rolling shutter, and if you are just doing stills, the dynamic range on these sensors can go up from 11/11.5 to about 13-15. Fairchild Semiconductor developed an sCMOS sensor, similar to the one used in the BMCC, it is 5.5MP and can output 100fps, very high DR, and very, very, low noise.

The issue isn't the sensor read out, it is the processing system, but not the initial processing, as debayering and chroma sampling aren't all that processor intensive. Compression is, and especially if you want high quality, and high speed, as you get into systems that need multiple simultaneous processing engines, and it gets expensive to develop!

It'll drop though, and not that far off either. Especially with sensor companies like CMOSIS, and Fairchild pushing relatively decent image sensors at low costs, and companies developing camera solutions using FPGAs instead of custom ASICs and general purpose DSPs like TIs Keystone DSP multicore system.

I'm not saying there are no fast sensors out there. I'm saying there are design tradeoffs: if you want your sensor to be fast enough for full-readout video, it will be either more expensive, or worse in some other dimension.

I know about the CMV12000, I've actually played with one. It's the perfect example of what I'm saying: speed and global shutter were their main priorities, and they didn't want to make it super-expensive (just quite expensive), and the result is that it tops out at 10 stops of DR (and you'll be hard pressed to find someone who considers 8 of those usable). The HDR mode can bring that to 13 but it creates motion issues and is not the best solution for video.
http://www.dvxuser.com/V6/showthread.php?299568-Question-about-RED-s-HDRx-why-don-t-people-use-it

"companies developing camera solutions using FPGAs instead of custom ASICs"
Yep, that would be me


PS: to join this discussion with the main topic here, what I'm saying is that the sensor in the 1DX and 1DC was designed with readout speed as a main priority, whereas the sensors in most other DSLRs were not. You can't do, with the sensor in most DSLRs, what you can do with the 1DC.