Concerning 3D filmmaking/editing

[JFTS]

This question is perhaps more about shooting than editing but since the latter plays a big role in 3D video production, I figured this would be a good place to ask.

It is known that for a 3D film to have optimal image quality, it has to be shot natively with two cameras on a 3D rig and that 3D post-conversions are generally inferior since there are scenes that depth cannot be simulated properly. A pinnacle of 3D cinematography is obviously the Hobbit trilogy (be it 48 or 24 FPS) and one of the worst 3D conversions I've seen was done on the first Thor movie. I am just writing these examples purely for reference and not about their quality as films.

However, I recently watched Mad Max: Fury Road on cinema in 3D and there was something very interesting about it. I read in an article that for financial reasons, the movie was shot in 2D (using Arri Alexa cameras) but with the intention to be converted in 3D. In other words the concept was a 3D film from the very beginning. And the result was very spectacular indeed!

On the other hand, I watched Transformers: Age of Extinction last summer which was shot on a 3D rig and it was absolutely terrible. The ridiculous camera panning of Michael Bay didn't help the situation. The image was moving too fast and it was exhausting for the eyes.

So immiediately we have two exceptions to the general rule I stated at the beginning of this post and it is an ideal comparison because both are action films which rely on similar shooting techniques.

But I simply cannot praise Mad Max's cinematography enough as a final result, which brings me to the question: How do you have to change your approach when shooting a stereoscopic film? I mean for example, avoiding very fast camera movements, being careful with the convergence point on closeups, having in your mind how each scene will look in 3D and then make adjustments according to that etc.

Again, Transformers proved that two cameras is not enough to get a good picture, while Mad Max (a film that is literally in constant movement) is one of the best 3D movies I have ever seen! And it was shot in 2D! How can such a result be achieved?

It would be great if I could get more information.

[smrpix]

You may be a bit early (there's no articles in Post Magazine yet.) It appears going 2D was primarily a financial/pragmatic decision. In a film like Mad Max because of all the motion, particularly horizontal motion, there is plenty of z-axis information that can be extracted simply by comparing sequential frames. Think of frame 1 = left eye depth data, frame 2 = right eye depth data, add computer, touch up.

[JFTS]

You may be a bit early (there's no articles in Post Magazine yet.)

You are right. I've probably made a mistake but I've definately read/heard it somewhere.

It appears going 2D was primarily a financial/pragmatic decision. In a film like Mad Max because of all the motion, particularly horizontal motion, there is plenty of z-axis information that can be extracted simply by comparing sequential frames. Think of frame 1 = left eye depth data, frame 2 = right eye depth data, add computer, touch up.

Can you explain this concept a bit more? I'm basically trying to figure out why post-conversion worked for this particular film and not for others.

[Cornucopia]

Sorry I didn't reply earlier (on vacation)...

I cannot comment directly on your evaluations of those 2 movies (Transf. & MadMax), as I haven't seen either one (and won't comment because I don't want to falsely presume), but I can speak in general as regards the creation of S3D media. (Not surprised, though, that strong panning could lead to problems)

S3D media these days falls into one or more of 3 categories: Natively-shot S3D, Shot 2D+Converted, and S3D-rendered 3D CGI. Nearly all of the animated properties are that last one (exceptions being those where it is logistically & financially prohibitive to revisit the rendering/compositing/finishing pipeline). Most live action, especially your adventure/scifi types, do not get created solely from the former 2, but rather include a sizeable and possibly substantial component of 3D CGI that is also rendered in S3D. Particularly popular is the use of CGI'd middle- and back-ground planes (with optimized cam perspectives for each) composited with greenscreeened fore-ground actors. You didn't list this type of setup in your original assessment, and it should not be discounted.

Now, there are LOTS of things that affect the quality of the S3D as seen in the theatre: depth grading/budget (and it's relation to the edit), amount and direction of parallax, comfortability of view (<divergence, based on screen size + distance + viewer age + pixel parallax), framerate & resolution of the chain, the use (or not) of non-depth related parallax differences...

For those titles that are shot in native S3D, there are vertical parallax issues that can crop up with toed-in shots, and FG/BG convergence/divergence balancing issues that arise with parallel shots. Then there's uncorrected registration errors.

For those titles that are shot in 2D+converted, the quality of the conversion is a mixture of influences: the skills & toolsets of the converting contractors, the time involved (usually because of budget), the depth priority (in-window, out-of-window), the proper setup (and possible floating) of the stereo window, etc...

Since there is no single defining gauge to the S3D quality, there can also be many ways in which each title can be positively or negatively assessed.

On how to make movies, one of the first places for a new S3D enthusiast to start learning is by reading "3d movie making" by bernard mendiburu. It's a few years old now, but still covers most of what needs to be understood about the process.

Regarding that suggested conversion method: motion parallax is a well-known phenomenon and it is fairly straightforward to transform motion parallax into depth parallax. There are other monocular dimensional phenomena that can be utilized: perspective & horizon line (including size change), texture gradient & discontinuity, and occlusion. These are all used to improve the guess of object+plane boundaries and in the calculation of the necessary HIT (horizontal image translation) to provide the accompanying (expected) binocular disparity. Then it's mainly a matter of object selection, HIT, and inpainting (to fill-in newly revealed occluded areas). Lather, rinse, repeat.

One area that Mendiburu and a lot of other S3D proponents ignore/downplay about S3D is in the area of those non-depth related parallax differences. In fact, most of the time, great effort is made to minimize or even eliminate ALL but depth-related parallax differences. What I'm referring to has to do with, for example, how BRIGHT a point is and it's comparative difference of view between the 2 eyes. Often GLARE is the most common incidence of this disparity, but all particulate effects and unusual texture effects can enhance S3D and make it "scintillating" and "vibrant" beyond what can be described just with horizontal displacement. Take a look at water/waves and it's "flickering" of reflections: this is impossible to duplicate with either the best quality standard 2D nor a 2D converted into S3D through only HT, and this is an area that I have been championing (both in my discussions on the subject and in the choice of S3D subject matter in my personal portfolio).

Some food for thought...

Scott