Zoom Lenses for the T2

[Jon Fairhurst]

As fasr as inspiration goes, it depends on the look that you want. If sharpness and clarity are the goal, then top, accurate glass will inspire. On the other hand, there are some old lenses that have a soft, high-character look. If you like that character, those lenses can inspire too.

Back in the pre-manual days of the 5D2, we shot a festival film with three old lenses that cost under $200 combined. It's a stylized film noir sendup, so those lenses - plus careful, low-saturation grading and lots of glow made for a very interesting look. For closeups, we used some cheap Hoya closeup lenses screwed onto an AI 50/1.8. It was far from sharp, and really added to the b-movie feel.

I'm not saying that crummy lenses are the goal, but the sharpest possible lenses aren't always the right answer. In fact, most recently I've been shooting with Glimmerglass filters to help knock down the highlights and add a slight glow. I need to add back a bit of contrast in the camera, but I like the look.

But if your goal is a super crisp look, get a super crisp lens and don't use diffusion. It all depends on your target look.

[Paul Curtis]

I would also add that in real terms, beyond a certain point, it's difficult to quantify sharpness in real images, especially moving (which introduces subtle motion blur into the mix). For example is the performance of a canon L 24mm lens going to be apparent compared to a cheap 24mm at video resolutions? Sure the look will be different, different bokeh and colour and better performance (less aberrations) but in terms of sharpness i think you'd be hard pushed to see a real world difference.

The L glass is lovely for stills, so if you're shooting both it's worth it, but for video only it seems overkill right now.

cheers
paul

[Erik Andersen]

Paul, do you have a 24 1.4? I do and it is breathtaking. Even compared to other L-series lenses. Tack sharp with gorgeous bokeh. It's so odd that L-lenses are routinely bashed and yet professional canon shooters uniformly use them.

The more unsharp the lens the better? Don't know what to say. If you start with an unsharp image, you can never put back the sharpness. But you can always blur a sharp image. In relation to aliasing, an approach suggested on another thread is to mask areas of the image that show aliasing artifacts and retain the sharpness elsewhere.

If the camera resolution is so terrible - it's true it's bad compared to, say, the EX3 - why would people be hacking the 5D2 and 7D to mount PL lenses on them? That's "overkill" for sure, but these are people that probably know what they're doing.

Modified 7D-PL mount camera | Philip Bloom

[Paul Curtis]

Yes, the 24mm is breathtaking, i've rented before but couldn't justify buying it as i have the 10-22 and i don't spend much time at 22/24mm.

Let me elaborate because i never intended implying a blank statement that unsharp lenses are better. I'm saying that the line skipping in these cameras create a void between pixels that results in big aliasing problems. Normally a sensor would have a filter in front of it to cut out the detail that would result in these artefacts. The 550D has one, but it's designed for full res stills, not 1080. Therefore this camera has effectively no low pass filter which means high details are going to be aliased. The line skipping makes it even worse because the gap between lines is so great hence mostly horizontal problems - you would need a filter that blurred more vertically than horizontally anyway. There are some great charts and videos around that show this.

For video work the lens needs to take on the role of this filter that is missing (or more accurately not high enough in the vertical direction). It needs to stop high detail getting to the sensor. It's a trade off, images will look softer (because they are) but remember that half the time the perceived resolution is fake, and when that is moving it looks terrible because this fake resolution changes from frame to frame hence the shimmer. Be it something with horizontal lines moving through the frame or fine detailing of brickwork in the distance. There are the caprock filters that do just that - blur the image.

My point is, stick a caprock on a 24L then you may as well stick a nice old lens on there.

The cameras are being modded with PL mounts for the simple reason that it enables cinematographers to use the same physical lenses they use to take stills with for grading and production purposes and perhaps some crash cam style shots. Sure, these cameras are being used in some cases theatrically, usually (hopefully) when there is a compelling reason for that form factor. Certainly not because of quality.

The issue is that there is a finite amount of detail any sensor can resolve. If you push too much detail to the sensor then you will create artefacts. The whole camera pipeline is carefully balanced bearing in mind the performance across all components. These cameras are primarily designed for stills. Somethings got to give and that is video resolution.

I hope that makes my statements a bit clearer?

cheers
paul

[Erik Andersen]

Thanks Paul. Are you able to point me to some videos that show the approach you've described? I've never tried it and as you can tell I'm skeptical! There's definitely a limit to how much the image processor and the codec can handle. But to my eyes sharp = good.

As to the PL mount, I've mostly heard of Arri and Panavision lenses being used. Not a crash cam in that case! Seems like the 7D is sometimes a good choice as a B or C cam to a much better A cam.

Then of course there are the mods enabling uncompressed out.

[Paul Curtis]

Erik,

There is a very good analysis and demonstration by barry green on the other video forum, i wouldn't want to link from here. Also Stuart at the prolost blog has some good background.

In terms of video:

YouTube - Moire test 5D MKII

Is great example of how noise is created. The netting on the curtain will cause different false detail at different zoom stages.

Even the new footage from the 550D posted by James Miller shows the typical type of artefacts in the daylight examples (brickwork, roof tiles and horizontal lines).

There are some tests with the caprock filters on vimeo too, although i don't have the links here. In fact i think it's dvinfo that they came from in the first place.

The long and the short of it is that there is no way to fix this without substantially blurring the image coming in.

It like looking through a window with venetian blinds and where the blind is you cannot see what is behind that so you can miss details smaller than a blind height.

You can be careful what you point at. You can stick to shallow DOF to blur detail and you can attempt to minimise in post. Sometimes you don't know you've got problems until you see footage, it's not always visible on the LCD.

I would say that for wide deep focus shots the cameras are next to useless but then there are plenty of cameras that are great at this. For 100mm shallow DOF MCU then they can look great but so can an EX1 at those kind of focal lengths. The sweet spot is really something like f1.4 at 24mm (for example) or 30mm where you can get a cinematic field of view with shallow DOF. In my opinion that's what the cameras are most useful for.

Also on the compression, the compression likes areas that are blurred. If you shoot someone in front of trees with deep DOF then the compression would fall apart. If you reduced the DOF then the high frequency detail of those tress would become blurred and compress easily, giving more 'room' to compress the person better.

Also i've had limited success on full res source hiding the moireing by simply adding film grain, which is more acceptable and can hide the underlying shimmering.

hope this is useful in some way
paul

[Jon Fairhurst]

The science is very clear. To eliminate aliasing, you need to cut out the high frequencies before they hit the sensor. And to keep things sharp, you want to preserve all the detail the sensor can handle, but not one line more.

For instance, at 1080p, one can show 540 black lines separated by 540 white lines at the absolute maximum. Ideally, we would use an optical filter that would show 539 lines cleanly, but would blur 541 lines to gray. Unfortunately, no optical filter has that sharp a cutoff.

RED really has it right. By shooting at 3K (wide), the filter can kill everything above 3K and pass everything at 2K (or 2.4K or whatever). Downsample digitally, and they should be able to deliver stunningly sharp 1080p with no visible aliasing.

Back to DvSLR lenses, everything that the lens throws away over 1080p is a good thing. Everything the lens throws away below 1080p is a bad thing.

But that's just resolution. Flare, bokeh, falloff, color rendition, focus distance, and focus accuracy/smoothness and CA are also important. To me those are the reasons to pay more for a good lens for DvSLR video. You don't need the most expensive lens to resolve 540 alternating lines.

[Paul Curtis]

Originally Posted by Jon Fairhurst

The science is very clear. To eliminate aliasing, you need to cut out the high frequencies before they hit the sensor. And to keep things sharp, you want to preserve all the detail the sensor can handle, but not one line more.

This is very true, except with these canons the problem is compounded by the line skipping. It means that the effective sensor 'pixels' have large horizontal gaps between them, most sensors the pixels would butt up next to each other. The effective resolution would probably be lower than 540 vertically.

Does that make sense or am i thinking incorrectly?

The aliasing isn't anywhere near as bad horizontally, so i guess entire lines would be read out and scaled correctly. Any aliasing visible would be down just to the OLPF in that case.

cheers
paul

[Fei Meng]

Originally Posted by Paul Curtis

The sweet spot is really something like f1.4 at 24mm (for example) or 30mm where you can get a cinematic field of view with shallow DOF. In my opinion that's what the cameras are most useful for.

I'm sorry to nitpick, but I gotta say that this is not quite right. In 35mm cinematography, 24mm to 35mm are considered wide angle. And f/1.4 is almost never used. The sweet spot is f/2.8 or f/3.5.

[Fei Meng]

Originally Posted by Erik Andersen

But to my eyes sharp = good.

Barry Green proved that the sharpness actually just an illusion due to aliasing. There is no more real detail/resolution with a super-sharp lens than with a softer lens.

The truth (and some people here might not like to hear/read this) is that if you want to achieve the sharpness of a super-sharp lens like a Canon L or Zeiss ZE, then just use a cheaper, softer lens and apply sharpening in post.

[Fei Meng]

Originally Posted by Erik Andersen

If the camera resolution is so terrible - it's true it's bad compared to, say, the EX3 - why would people be hacking the 5D2 and 7D to mount PL lenses on them? That's "overkill" for sure, but these are people that probably know what they're doing.

Sharpness and resolution are not the only reasons to use a top-class lens like a PL. PL lenses are designed specifically for cinematography. Not only do they offer superior contrast and color reproduction, but they also offer superior build quality and mechanics, optimized for cinematography. For example, the focus throws are very long, and the aperture rings don't have click-stops.

[Erik Andersen]

Originally Posted by Fei Meng

Barry Green proved that the sharpness actually just an illusion due to aliasing. There is no more real detail/resolution with a super-sharp lens than with a softer lens.

In my experience, aside from aliasing (which in most shallow DOF shots is minimal), it's a pretty good illusion if it is just an illusion. Could you post a link to the "proof"?

[Jon Fairhurst]

Originally Posted by Fei Meng

Barry Green proved that the sharpness actually just an illusion due to aliasing. There is no more real detail/resolution with a super-sharp lens than with a softer lens.

It's possible that a soft lens will not just lose resolution in the aliasing region (that's good), but also lose it in the visible region (that's bad). Also, if the visible region has problems in the visible band that vary by bandwidth (color), that could show up in the video.

I still come back to the target look. If you want a dreamy feel, super sharpness might not matter. Corner sharpness matters little if you shoot shallow DOF with a centered subject. On the other hand, if you're doing travel promotion videos, corner to corner sharpness could be important.

[Fei Meng]

Originally Posted by Erik Andersen

In my experience, aside from aliasing (which in most shallow DOF shots is minimal), it's a pretty good illusion if it is just an illusion. Could you post a link to the "proof"?

Aliasing plagues all video images from all DSLRs, even when it doesn't look like aliasing. And that's the most disconcerting part.

Start reading a third of the way down the page here: "Aliasing" by Barry Green

[Fei Meng]

Originally Posted by Jon Fairhurst

It's possible that a soft lens will not just lose resolution in the aliasing region (that's good), but also lose it in the visible region (that's bad). Also, if the visible region has problems in the visible band that vary by bandwidth (color), that could show up in the video.

I still come back to the target look. If you want a dreamy feel, super sharpness might not matter. Corner sharpness matters little if you shoot shallow DOF with a centered subject. On the other hand, if you're doing travel promotion videos, corner to corner sharpness could be important.

To clarify, when I said "softer lens," I didn't mean a soft lens. I only meant a lens that was not super-sharp.

One important about corner-to-corner sharpness that must be said is that it doesn't matter very much if you're using lenses intended for full-frame or 35mm cameras on a crop-sensor camera. The cropped sensor only uses the sharpest region of the lens, which is the center.