Breaking the codec. Why the 5D isn't quite ready for primetime

[Perrone Ford]

So as some are aware, I was involved in shooting a short film back in October and we are now going through post. I've watched these 5D and 7D forums intently because we shot the short on the 5D. There have been numerous posts about how good the 5D is, and how robust the codec is, etc.

I've been waiting to get to this part of the post-production process to comment here. One of our scenes was a tracking shot of a young woman running through the woods. We've got trees in the foreground passing between us and the woman, she's really running, and we are on a stabilizer tracking with her. Tough for any codec. So I thought I'd post some stills from that for everyone to analyze. If anyone wants the DPX files, I can post that too.

I am not saying the camera isn't good. I think I've talked myself into a 7D next year. But it's these kinds of things like I am displaying here that will cause issues for a major production, or even a small production with high motion.

Pay attention to the red handbag and it's edges, the black portion of her dress, and the area of her neck around the ears. This was shot in full sunlight, so this is not a situation of the camera being starved for light. It is merely starved for bits.

Comments welcome.

[Chris Barcellos]

1. What was shutter speed. What was ISO setting. Was there an ND filter employed.

2. It is hard to judge what complaint is with three stills obviously separated by several frames. I do see detail disappearing in places, but how does the final sequence appear. I think you can take any film apart frame by frame and get similar complaints, depending on motion, action, and attendant blur. In one frame, for instance the purse in a swinging motion has been caught at a point where it is relatively stationary, in another, it is obviously in motion relative to the camera.

[Reggie Moser]

Dude are you kidding me, the 5d is the best thing since slice bread for indie filmmakers at that price range......... test test test before your begin production that way you can figure out any issues you may need to work around....last i check all cameras have their own issues.......look at all the things you have to work around with the Red cams and those prices are thru the roof.......... I'm really just messing around having fun so don't take what I'm saying for serious....but really the 5D changed my life and inspired me all over again.....happy shooting

[Bill Davis]

The next time you shoot something like this, you should actually fire a confetti canon at some point during the scene. And make sure it's multi-color confetti. That way you would actually get even closer to achieving the single "most difficult to compress" shot ever attempted in a digital medium.

Let's see, a forrest of leaves and mottled sunlight working hard to make sure that each and every pixel shifts both color and luminance value between every frame - CHECK. A fast moving central figure constantly in fast motion - CHECK. Now all you need is the confetti.

You BET your result is "BIT STARVED." And you starved it by attempting to feed it a data throughput stream that is WAY TO HUGE to slide down it's gullet, dude.

Seriously, part of the reason that NTSC was so easy to make look decent, was that the resolution was awful and the motion interpolation was marginal at best - particularly since the image was interlaced which further sacrificed picture resolution for better temporal resolution. So SD TV is more like a fast series of impressionist paintings than a series of sharp images. Anyone who's ever tried to pull a still off SD video understands this all too well.

Here's the reality. It's HARDER to get good rapid motion video in high def then it is in standard def. This is because grabbing and processing those big-assed frames that everyone seems to want is a major throughput and storage hassle.

HDTV in 2009 is a series of COMPROMISES. One of those is about rendering fast motion in complex scenes.

The experienced shooter understands this - and writes and blocks scenes to avoid things the camera can't do well. Like this kind of scene.

Take the same girl running - move the camera so that she runs at an shallow angle toward it and/or away from it, with the camera fixed so that the trees and ground cover are STATIC. Push IN for more close ups using the 5dMkii's exceptional depth of field to blur out the background when it's in motion. Use swish pans and fragmented shots to imply speed and progress. All of those will compress a LOT more efficiently than a pan that jiggles every pixel on every frame refresh.

Hope this doesn't sound overly harsh - but you have GOT to understand the nature of how your tools actually work when they get this complex, or you're going to open yourself up to poor results.

Simple as that.

[Perrone Ford]

I knew this was going to be a fun one...

So let's see.

1. I do not know what the ISO and shutter speed were for certain. But based on the metering, I recommended we shoot at ISO 100 and I think that was adhered to. No ND was turned on. Shutter speed was likely 1/48 which we knew would get us motion blur. That was expected.

2. The separation of the stills by several frames is not the issue at hand. For me, it is the blocking up in the blacks. Actually, I am quite pleased with how the foliage came out. Film doesn't block like this. This is all digital.

3. @Reggie. No dude, I am not kidding you. :) Yes, the 5D is a wonderful tool for indie filmmakers. Hence why it was purchased and used to shoot this project. I'd say for 95% of the project I was extremely pleased with the output and have posted numerous stunning still here from the camera. Some graded, but most not. I am not disparaging the camera whatsoever. Merely pointing out that on heavy action, even while tracking, it will break up, and that might affect how someone uses the camera.

4. @Bill. I didn't direct this project, I didn't block it, or design the shots. I was merely tasked with lighting it, and trying to get the image into the camera. I agree with you about blocking and trying to ensure good results based on the camera in question. This kind of shot would test a great many cameras.

The point of posting these stills is not to disparage the Canon 5D. It's a wonderful camera for the money. The point is to show that it has limits, that we found them and stepped over them, and to show the results of doing so. Nothing more, nothing less. I did a similar test with the DVX100, and with my EX1. And believe me, I am VERY happy with both those cameras.

[Chris Barcellos]

It does seem like this codec blocks the darks hues badly at time. I think shooting with a flat shooting setting like super flat will help resolve some of that. What picture style setting were you using ?

I am attaching a spicture style that is designated as superflat. You will need to change extension to .pf2 and then upload to one of the user picture stlyes through the eos software. Test that and see if it works to alleviate issues. You can add sharpness and levels as needed in post.

[Perrone Ford]

Again, I was gaffer and AD. Not my camera, I don't have access to it, and I only know what information I relayed to the camera operator. I have no idea what profile, or anything else was used.

When I shoot, I tend to shoot quite flat. Many people tend to try to get their looks in the camera even when there is going to be an extensive post. I don't do that personally.

Do you think shooting superflat would have prevented some of the blocking up in the blacks and on the skin tones?

[Perrone Ford]

For those who want to view the clip...

YouTube - Stabilization test of tracking shot through the woods.

[Jay Houser]

I have a question:

The limitations of DSLRs is in the speed of the sensor in reading a frame and the ability of the chips to process all of the information. Right?

Wouldn't the addition of 2 more equally sized sensors and a few more chips (like a real video camera) be more capable of keeping up?

I realize this would require a large video camera form factor and hopefully not larger, more expensive lenses.

Hopefully we will see something along these lines coming from Canon. Or Sony, Panasonic.

[Chris Barcellos]

Originally Posted by Perrone Ford

Do you think shooting superflat would have prevented some of the blocking up in the blacks and on the skin tones?

The way I look at it, the less you ask the codec to do during intial shoot/capture, the less potential there is for the codec breaking down. Shooting flat seems to take out sharpness and other things that I can add later after to converting to Cineform and using the 4:2:2 format that is created.

[Perrone Ford]

Interesting comment. I've not found that shooting flat is any less taxing on the codec from a luminance perspective. Protecting the shadows and highlights pays dividends in what you come home with but not sure it helps the codec. Scaling back the colors certainly seems to keep things in check and allows for easier coloring in post, but I can't see how that would give the codec more bandwidth to use.

Maybe I just don't understand how it all works.

[Chris Barcellos]

Originally Posted by Perrone Ford

For those who want to view the clip...

YouTube - Stabilization test of tracking shot through the woods.

Were your grabs from original footage, or from the post stabilized material ?

What is stabilizing being done with DeShaker with virtual dub ? That was a pretty shakey pre stablizing shot, and not so sure the codec should bear the blame.

Again, I do think that converting the material to something like Cineform helps in running through a stablizer in post.

[Chris Barcellos]

Originally Posted by Perrone Ford

Maybe I just don't understand how it all works.

You can't pin knowing how a codec really works on me.... I am just rolling along trying to find what provides best results..:)

[Perrone Ford]

Originally Posted by Chris Barcellos

Were your grabs from original footage, or from the post stabilized material ?\

Original footage.

Originally Posted by Chris Barcellos

What is stabilizing being done with DeShaker with virtual dub ? That was a pretty shakey pre stablizing shot, and not so sure the codec should bear the blame.

Yes, footage went through VirtualDub. Since VDub won't take 5D footage, it was converted to the lossles Lagarith AVI codec.

Originally Posted by Chris Barcellos

Again, I do think that converting the material to something like Cineform helps in running through a stablizer in post.

Agreed. And that's how I do it. I don't edit long-GOP on the timeline.

[Cris Daniels]

ProLost - ProLost Blog - You Didn't BelieveMe

as much as I like the 5D and 7D and shoot them almost every day, this article will explain why a real video camera does it better than a still camera with video capability