hdslr video vs film

[VegasBud]

Considering that VCDHelp (aka VideoHelp) hosted quite a few discussions of dvd technology when vcd was the standard of the time, I would have thought there would already be a thread at VideoHelp concerning the "Zacuto Camera Shootout 2010", but if there is such a thread, I can't find it...which is why I'm starting this one. Even if the subject doesn't directly impact the average user of these forums, at this point in time, it seems likely it will influence how you (and the studios) do things in the future.

In a nutshell, the "shootout" is a series of tests comparing hdslr video and traditional 35mm film (Kodak and Fuji, 2k and 4k scans).

The "shootout" is located here. There are three videos at that webpage, each covering different areas, so it's best to watch all three. Some material develops as you work your way through the parts, so viewing the parts in sequence would probably be a good idea. Each part is about a half hour runtime, so it takes a little more than an hour and a half to watch the entire thing.

Ken Rockwell (on his site) cuts to the chase much more concisely than I can, so I'll just quote a part of what he said:

Holy guacamole, amigo, I've just seen the best comparison ever. Set aside a half hour, dim the lights, and sit back and enjoy the movie.

This isn't some guy in his mom's basement taking pictures of magazines.

This is a Hollywood-level comparison of DSLR video versus 35mm motion picture film, done by people who know, like the ASC members who shoot all Hollywood's movies.

The purpose of this is to see the differences between 35mm motion-picture film and DSLR video for professional motion picture production.

I should point out that since Mr. Rockwell says "Set aside a half hour...and enjoy the movie", he appears to have only watched one of the three videos, and may have missed 2/3 of what was actually available. Hopefully, VideoHelp viewers won't have the same oversight.

By the way, this subject isn't some wild jaunt into "what if" science fiction. Lucasfilm appears to be seriously considering integrating the technology in some new movie they're making (I think it's about the Tuskegee airmen?). Also the season finale of "House" (this season) was shot entirely using hdslr video. Unfortunately, I don't watch "House", so I didn't get to see it on a hd tv, but I quizzed the heck out of my sister (a "House" fan), and she didn't notice anything unusual about the episode's video quality at all. Then again, like most viewers, (to her) the technology is only a means to an end, and it's the story she's interested in. Still...if shooting with hdslr didn't interfere with the storytelling, that's a statement in itself.

One of the series of tests I found particularly interesting was about shooting at high ISO using light from a bic lighter. The Nikon D3s, which was handicapped in the rest of the tests by its' 720p limit, was quite impressive, even at 10,000 ISO.

If someone does have a Canon EOS 5D Mark II, here's a link to a video showing Philip Bloom (the current hdslr guru) showing the menu settings to use for optimum video use.

Here's a link to the video Philip Bloom shot at Skywalker Ranch (as a demo for Lucasfilm).

Here's a link to the page at Ken Rockwell's site where he comments on the "shootout".

For more information, google using "hdslr" (without the quotes). Add whatever qualifiers you want to zoom in on specific aspects...for example, googling "hdslr sound" (without the quotes) will open the door to whatever you want to know about the subject.

On a somewhat connected topic...in the Ken Rockwell review of the "shootout" he says:

...remember that 35mm movie film is only half the size of 35mm still film. In this test, the film frames are only 18 x 24mm, not 24 x 36mm...

While Mr. Rockwell is generally very precise with technological facts, that surprised the heck out of me. Just to be absolutely sure (in my own mind, if nothing else), does anyone have experience with 35mm movie film...so as to confirm or disagree? I would appreciate it.

[jagabo]

On a somewhat connected topic...in the Ken Rockwell review of the "shootout" he says:

...remember that 35mm movie film is only half the size of 35mm still film. In this test, the film frames are only 18 x 24mm, not 24 x 36mm...

While Mr. Rockwell is generally very precise with technological facts, that surprised the heck out of me. Just to be absolutely sure (in my own mind, if nothing else), does anyone have experience with 35mm movie film...so as to confirm or disagree? I would appreciate it.

I never thought about it before but I believe he is right. In both cases the film is 35mm wide (from edge to edge, including the sprocket holes). But in a movie camera the landscape picture is recorded horizontally across the film. There's only 24mm of space between the sprocket holes so you get a 24x18mm picture (even less when there's an audio channel). In a still camera the the landscape picture is recorded vertically along the length of the film. So you get a 35x24mm picture.

35mm movie film: http://en.wikipedia.org/wiki/35_mm_film
35mm camera film: http://en.wikipedia.org/wiki/135_film

By the way, thanks for the heads up on the comparison. I haven't watched all of it yet.

But near the end of part 3, where they're simulating raw output from the DSLRs (via pass through?) it appears they've made a very amateurish mistake. They are seeing more shades of gray in the raw caps than in the h.264 caps (from the flash cards). The differences they are seeing seem only to be matter of PC.601 (raw) and REC.601 (h.264) decoding. If they had used PC.601 decoding on the h.264 caps they would have seen just as much dark and light detail.

[VegasBud]

jagabo,

Thank you for that. I appreciate it.

[jagabo]

I added a bit to my earlier post...

I think they also made a mistake with rec.601 vs rec.709 decoding of the h.264 video.

[edDV]

Very interesting. The results vs. film were as expected. To close the gap, one needs to avoid the h.264 codec during acquisition (e.g. SDI SMPTE-292M) and add bit depth. 8 bits just isn't going to cut it for acquisition (pre color correction). Listen to the colorist's complaints.

When you add a "raw" (less compressed) recording format and 10-14 bits per component these become very similar to the Red or other HD cinema cameras. Only the form factor and lens/sensor scale differs.

I would have liked more motion studies to show effects of h.264 compression on resolution and artifacts. The tests used mostly low motion static shots where a still cam will do well but the model movement in the green screen test was helpful. I'd like to see this again on a HD tape to a good large screen.

[jagabo]

To close the gap, one needs to avoid the h.264 codec during acquisition (e.g. SDI SMPTE-292M) and add bit depth. 8 bits just isn't going to cut it for acquisition (pre color correction). Listen to the colorist's complaints.

Yes, capturing raw with more bit depth (or some lossless or less compressed format) would give cleaner results. But I think a large part of the problem was in their processing of the h.264 video. It looks to me like they used a rec601 matrix to convert the h.264 to RGB but used a pc.601 matrix to convert the raw video to RGB (or maybe it comes out of the camera as RGB?). I think that is why they were seeing darker darks and brighter brights in the h.264 video.

I'll have to watch again but as I recall there was talk of h.264 encoding as RGB. As you know h.264 uses YV12 internally. So I suspect the h.264 decoder they were using was putting out RGB and performing the rec.601 contrast enhancement. (edit: My memory was correct, at 19:40 in the third video they say h.264 is RGB and rec.601.)

I took some screen grabs of the grayscale bars they used to demonstrate how the raw caps contains more dark detail. Here are the crops from the images they posted, h.264 top, raw bottom:



While showing these two samples they were raving about how the raw video showed much more dark detail than the h.264 video. But here I've adjusted the raw image to use rec.601 conversion rather than pc.601:



Maybe it's just coincidence that the differences almost exactly match the difference between rec.601 and pc.601 conversion (note that those scenes were from a camera pointed at a monitor so it's not surprising that my adjustments don't match 100 percent). But it looks suspicious to me. I wish they would post some sample video files.

Similar brightness differences can be seen in the later shot with the woman on a brown chair. There is also a color shift there which I think may have come from using a rec.601 matrix rather than a rec.709 matrix. I recall them complaining that the h.264 video was rec.601 -- which probably isn't true since it was HD.

Here are the levels charts they provided for the gray bars. h.264 top, raw bottom:





Again, this look almost exactly the the difference between rec.601 and pc.601 YUV to RGB conversion.

Maybe they knew exactly what they were doing and just used this as a shorthand simulation (because it would be impossible to see the difference between a 12 bit cap and an 8 bit cap in a VP6 FLV file posted on the web) of how greater bit depth and less compression would give you more to work with? If so, they should have explicitly said this. And it doesn't make sense that they would show the levels graphs it that was the case.

[poisondeathray]

Very interesting. The results vs. film were as expected. To close the gap, one needs to avoid the h.264 codec during acquisition (e.g. SDI SMPTE-292M) and add bit depth. 8 bits just isn't going to cut it for acquisition (pre color correction). Listen to the colorist's complaints.

When you add a "raw" (less compressed) recording format and 10-14 bits per component these become very similar to the Red or other HD cinema cameras. Only the form factor and lens/sensor scale differs.

I would have liked more motion studies to show effects of h.264 compression on resolution and artifacts. The tests used mostly low motion static shots where a still cam will do well but the model movement in the green screen test was helpful. I'd like to see this again on a HD tape to a good large screen.

One has to remember these are $1-3K cameras intended for still photographers, who occasionally might shoot a video bit. The actual market for Indy filmmakers and those that use the in big productions (for b-cam shots) is tiny compared to the photography market. (They actually shot an entire episode of "House" with the 5D)

How many of target market would be willing pay extra for those features, like HD-SDI out, RAW access, better implementation of codec ? I think a big part of the allure of these cameras is the low cost (compared to traditional setups). It would cost a fortune.

While interesting, they didn't even touch on the the achilles heel of these models, aliasing and moire, and CMOS skew. Some of it was evident in the blow up shots, but it happens much more frequently to not even give it a mention. Every "DSLR" video forum has posts riddled with these problems. It's quite difficult to get clean shots. You need to have handled these alot, and know there are severe limitations in the type of shots you can use. Now if it was possible to access the RAW data (before all the damage is done, the signal processing, and line skipping/sampling), that would be another story

But for the price, no one will argue that these are amazing cameras.

I'll have to watch again but as I recall there was talk of h.264 encoding as RGB. As you know h.264 uses YV12 internally. So I suspect the h.264 decoder they were using was putting out RGB and performing the rec.601 contrast enhancement.

Nice hypothesis Jagabo. I caught that RGB slip up too, it was the guy talking about ProRes (which is Y'CbCr 4:2:2). He spoke with such enthusiasm, but something didn't quite add up.

Phillip Bloom posts frequently on the various DSLR forums, and is very interactive. He answers lots of questions on the specifics of the tests (color profiles used, settings used) , but I doubt he would know about the last test and the decoding specifics. I haven't seen any of the colorists post on the forums