New idea for bigger resolution, devs - get in and check it out!

[Thejungle]

Hello all!

I am preparing my 600d for optimum options to get most of it for my future film. And as there was day like everyday, an idea came across my mind when I noticed the "anamorphic" option in ML menu.

Now what I think would be a good idea is to be able to record squeezed horizontaly RAW video. What makes horizontal squeezing better then vertical is - you don't loose so much quality, because there is no line skipping. So you get no aliasing. The only quality loss you could notice is when horizontally stretching the image, but if the ratio isn't too big (e.g. 1.2-1.3), you could handle any "artifacts" with resampling engines in RAW processing software (lightroom, Blow Up 3 etc). But! If we could save some pixels horizontally, maybe (!) you could just get this HD finally in 600d RAW! I'm definitely looking for it, I think it would be a really big improvement for our sad 21 MB/s buffer.


BTW. I wanted to mention this is not totally new feature as you still get vertical squeezing when using 720p 50fps option in Canon menu. So... this is quite good, but as I said - vertical squeezing destroys the image (aliasing), and horizontal - not so much! I hope we dould get it working, lovely idea. What you think guys?

Cheers!
-theJungle

[Midphase]

I don't understand what you're asking.

[Thejungle]

I don't understand what you're asking.

You are recording squeezed video: e.g. you're recording 1280x720 but squeezed to for example 1000x720. It gives you more buffer. You can stretch video in post to get 1280 horizontal res.

[tupp]

I am fairly sure that ML/TL users are already doing this.

As I understand it, in the instances in which it is possible to use ML/TL to change the aspect ratio to more of a 4:3 or square format, one can sometimes increase the actual vertical resolution (to higher than 720) while using anamorphic optics (oriented horizontally), so as to get a greater effective resolution overall.

[Thejungle]

I am fairly sure that ML/TL users are already doing this.

As I understand it, in the instances in which it is possible to use ML/TL to change the aspect ratio to more of a 4:3 or square format, one can sometimes increase the actual vertical resolution (to higher than 720) while using anamorphic optics (oriented horizontally), so as to get a greater effective resolution overall.

Yeah, but we are not talking about anamorphic lenses. It is just an idea for a little compression to get better resolution in effect.

[chmee]

There's no room for subtracting/cropping/dividing/skipping raw-data, because there's no "processing-unit" between sensor-output and memory-controller - thus, i'd say, not possible - but a1ex and the masters of hardware should tell us..

regards chmee

[Thejungle]

There's no room for subtracting/cropping/dividing/skipping raw-data, because there's no "processing-unit" between sensor-output and memory-controller - thus, i'd say, not possible - but a1ex and the masters of hardware should tell us..

regards chmee

It is possible, because you already can shoot with vertical squeezing in 720p 50fps setting, as I mentioned before.

[dmilligan]

you're recording 1280x720 but squeezed to for example 1000x720

I also have no idea what you mean, please define "horizontal squeezing" mathematically or with pseudo-code. How exactly do you get from 1280 to 1000? throw away some pixels (aka line skipping, or in this case 'column' skipping)? bilinear interpolation? magic?

It is possible, because you already can shoot with vertical squeezing in 720p 50fps setting, as I mentioned before.

I have never seen a "squeezing" option anywhere

[tupp]

I think that I am beginning to understand what you want.  "Squeezing" during shooting and "expanding" in post are misnomers.

What you actually seem to be proposing is to change the aspect ratio of the pixel so that there are fewer pixels on the horizontal axis.  This would create, effectively, rectangular pixels which are longer on the horizontal axis.

If so, there probably would have to be some sort of pixel binning or pixel skipping of the raw data done within the camera.

[Thejungle]

I think that I am beginning to understand what you want.  "Squeezing" during shooting and "expanding" in post are misnomers.

What you actually seem to be proposing is to change the aspect ratio of the pixel so that there are fewer pixels on the horizontal axis.  This would create, effectively, rectangular pixels which are longer on the horizontal axis.

If so, there probably would have to be some sort of pixel binning or pixel skipping of the raw data done within the camera.

That's it. I'm proposing to narrow images (yeah, pixel aspect ratio could be changed) to write them while recording, and to stretch them to proper aspect in post. I don't think about line skipping though, because it will of course destroy image. What I mean is to compres/resize them during recording as it is done while you record in 720p.

[tupp]

Just for clarification, there is no "resizing" of the image -- the full 16:9 area of the sensor is being utilized.

However, you are setting a horizontally oblong pixel aspect ratio in your camera, while setting the same pixel aspect ratio somewhere in post.

[Thejungle]

I also have no idea what you mean, please define "horizontal squeezing" mathematically or with pseudo-code. How exactly do you get from 1280 to 1000? throw away some pixels (aka line skipping, or in this case 'column' skipping)? bilinear interpolation? magic?
I have never seen a "squeezing" option anywhere

Open canon menu and change recording resolution to 720p 50fps. Now get back to ML menu -> recording -> raw video/mlv video -> resolution. Now you will get vertically lower resolution and info: "stretch vertically in post by 1.44 to get proper resolution of xxx".

However, you are setting a horizontally oblong pixel aspect ratio in your camera, while setting the same pixel aspect ratio somewhere in post.

You name it.

[tupp]

By the way, if this function is possible, moire' and aliasing could increase horizontally, especially if the solution is accomplished through pixel skipping.

[1%]

canon already did the squeezing for 720P.. they did it to get that 50/60 fps

[Midphase]

That's it. I'm proposing to narrow images (yeah, pixel aspect ratio could be changed) to write them while recording, and to stretch them to proper aspect in post. I don't think about line skipping though, because it will of course destroy image. What I mean is to compres/resize them during recording as it is done while you record in 720p.

If you record an image at 1000 horizontal pixels and then expand it in post to 1280 you're going to get distortion in the image (unless you use an anamorphic lens). If you want the sensor to somehow throw away a few lines here and there and give you a weird squeezed image that you can then desqueeze in post, then you'll get some pretty bad moire and aliasing.

I don't think it would work either way. Do yourself a favor and buy a Blackmagic Pocket Camera if you're really itching to get higher end raw video at a comparable cost....or stick to h.264 for the time being.

[tupp]

If you record an image at 1000 horizontal pixels and then expand it in post to 1280 you're going to get distortion in the image (unless you use an anamorphic lens).

That is not what Thejungle is proposing.

In his scenario, the pixel count on either axis never expands nor contracts from shooting through post.  He is merely proposing to use pseudo rectangular pixels, which are effectively longer on the horizontal axis (similar to some of the older Panasonic cameras, I think).

[dmilligan]

and how exactly do you propose to make the square pixels rectangular? magic?

[Canon eos m]

and how exactly do you propose to make the square pixels rectangular? magic?

lol! 

[tupp]

and how exactly do you propose to make the square pixels rectangular? magic?

Binning each axis at a different rate is one way, but it sounds like the OP is okay with "pseudo" rectangular, which would simply mean using only one pixel for every "n" pixels along one axis.

I do not know whether or not either method is possible with the EOS-M, but I would doubt it, after reading about the many significant limitations of Canon cameras.

Certainly, one can make the pixels rectangular after the footage is recorded, but the shooting bandwidth problem would remain.

Of course, there are  cameras/sensors with rectangular pixels, but that is not the case here.