How to Critique 3D

by Matthew DeJohn
©2010 MKPE Consulting LLC All rights reserved worldwide
originally published in the 15 April 2010 issue of Digital Cinema Report

With the flood of 3D content coming, no fewer than 25 films this year, it is important to understand how to critique 3D technically and artistically. What does good and bad 3D look like? What causes 3D to be bad or good? Admittedly, exposing some of the dirty secrets behind creating 3D may seem like a bad idea for someone working in the 3D business like myself. However, my hope is that in the current rush to get as much 3D on the screen as possible that education will help keep quality high. As opposed to broad brush critiques that hurt the 3D industry as a whole, focused and educated critiques will drive the industry to create the high quality work that we all want to see. To critique 3D content it is important to understand the three methods of creating 3D (CG, Stereo capture, and conversion), their technical issues, how to identify these issues, and what 3D should be achieving artistically. Once you're armed with this knowledge, you should be able to fully critique the massive amounts of 3D content coming at you.

CG Animation (CG-3D)

CG animation is arguably the most mature and refined mode of 3D content creation. Most of the 3D work coming out of this area is of very high quality. However, there are still some technical issues that can arise. As an overview, CG animation creates 3D by capturing a shot with a pair of virtual cameras. Also, multiple pairs of virtual cameras are often used to make most effective use of the available depth. For example, the amount of shape you perceive for a particular character may be manipulated independently from the rest of the elements in the shot.

Once the scene is rendered from these virtual cameras every element in the scene that has geometry will appear 3D. Since these cameras are essentially "perfect" and every parameter is controllable in the computer these shots should be technically perfect. However, many times elements are used that do not have geometry, such as matte paintings. That being the case, Matte paintings will appear flat unless steps are taken to remediate this issue (see 2D to 3D conversion). This issue is most easily spotted in stereo (with 3D glasses). The matte paintings in the shot may look falsely flat because you cannot perceive shape or separation where you would expect to.

Stereo Capture

The stereo capture approach is likely to be the most widely used. Stereo capture is achieved with a pair of cameras arranged in a side-by-side fashion or in a beam-splitter rig. The side-by-side arrangement aligns cameras similar to the way our eyes are aligned. But there are limits to how close the cameras can get together, and this limits the director's control of the depth of the scene. The beam-splitter rig alleviates this problem by allowing the cameras to capture perspectives that are closer than the cameras can physically get. One camera points forward while the other one is angled down, capturing the reflection of a scene. The core of technical issues that can arise with stereo capture arise from imperfect camera alignment, out of sync cameras, the fact one camera may be capturing a reflection of the scene, and the finality of stereo capture. Finality, meaning what you capture is largely what you get in terms of depth.

Improper image alignment is a common stereo capture issue that is caused by misaligned cameras or shots with uncorrected convergence. When viewing in 3D, this issue may make it difficult for your eyes to lock onto an object or it may cause you to get a headache. An easy way to identify alignment issues in the theater is to take off your glasses and try to spot an area of high contrast. You may see that a bright spot is a little higher for one eye than the other.

Exposure difference between the two cameras is another technical issue that can arise. This may be caused by an unmatched exposure or, more commonly, by the nature of the beam splitter rig. Because one camera in a beam splitter rig is capturing a reflection of the scene it will likely differ from the camera capturing the scene straight on. Exposure differences when viewed in 3D usually look like the image is glowing or has a sheen to it.

Bright objects such as glint off chrome, lens flares, and spotlights often create technical issues during stereo capture. Glint will often appear to be a different shape in each eye. This will make it hard for your eye to lock on to and it may just look messy. Lens flares often look odd in stereo because they are an effect specific to the lens construction. When captured they actually "invert" and sink into the rest of the scene. This usually looks confusing and takes the viewer out of the story. When the camera points at a spotlight the spotlight often has a bright star pattern at its source. In stereo the star pattern will often be rotated differently in each eye, making it hard to lock on to and uncomfortable to view.

A variety of sync issues may arise if the focus, zoom, or the shutters are not perfectly synced. In stereo, focal differences may make it feel like one lens of your glasses is smudged. If the zoom of the cameras are not synced it will be difficult to lock on to an element. If you take off your glasses you should see an image that almost looks like a radial blur. In this case, there will also be vertical alignment issues which are a large source of eye fatigue. If the shutters are out of sync, each camera captures the scene at a slightly different time. With a falling object it will appear farther down in one eye, creating a vertical alignment issue. With a horizontally moving object this sync issue will make the object appear to be floating or sinking. This issue may also make the image look stuttered.

Too much depth in the screen (beyond "infinity") is a much debated technical issue that can arise during stereo capture. "Infinity" is defined as the point at which elements in the left and right are separated by 2.5 inches (the width of adults' eyes) when projected. Some say this limit should never be exceeded since doing so means the viewer's eyes start to diverge, or go "wall-eyed". Others say there is a certain amount of flexibility to that rule or that it's not a big deal if the background is out of focus. It suffices to say that the point at which there is too much in-screen depth is the point where the viewer can't comfortably lock onto distant, in-focus objects. This issue can be identified by looking at the 3D image without glasses. Look for distant objects like mountains or spotlights; If the doubled-image of the mountain appears separated by many inches or feet the background is probably too far away.

2D to 3D Conversion

There has been a lot of chatter lately about the quality of 2D to 3D conversion. This criticism should be embraced and used as an impetus to improve, but it should not be misconstrued to suggest that 2D to 3D conversion has inherent quality deficits. Quality is controllable and many of the technical issues that can arise during conversion come from the lack of attention or scrutiny in the artistry of the technique.

The conversion process breaks down into 3 broad stages: roto/matte work, depth creation, and paint. During roto/matte work mattes are created for the major layers of depth. Usually, that means the layers have distinct separation in space. In simple CU's there may be a few as 3-4 major layers, but in more complex crowd shots there may be dozens of distinct layers. Depth creation can be achieved in a variety of ways, but it suffices to say that the goal is to create one or more new perspectives of the shot with the aid of geometry that approximates the objects in the scene. Paint work is necessary because the new perspective reveals background elements not captured by the camera. Background gaps must be filled with realistic imagery. Visual comfort can be affected when when any of these artistic stages are shortchanged.

A "rubber-sheet" effect is often observed when corners are cut to reduce the cost of conversion. This occurs when bypassing detailed matte and paint work, instead attempting to create depth with no distinct separation between elements. Without glasses it will appear that there is little separation between the left and right images, meaning the image will not look very blurry. In stereo these shots will appear unrealistic and the edges of foreground elements will wrap unnaturally back to the background, hence a "rubber-sheet" effect. In addition to this, smaller elements such as distant characters may reside at the same depth as closer characters. This approach usually does not achieve the expansive feel or accurate sense of the environment that is present in good 3D films.

Luminance-based 3D is another approach that can shortchange the viewer's perception of depth. This technique is often applied to elements such as flames or waves, bringing the lighter elements closer to the audience than the darker elements, or vice-versa. A 3D effect is created, but the results are often unrealistic and contradict the motion of that element. For example, the dark side of a wave may actually be closer than the light side or an explosion with smoke and fire may have dark and light shading that are actually at the same depth.

Other deficits in the depth process can be attributed to just plain inaccuracy. There are a variety of reasons for this, but most often it is due to the lack of detail when modeling scene depth, or inaccurate interpretation of the scene depth. If there is not enough detail in the model the scene will look overly simplistic and rough. If the interpretation of the depth is inaccurate the depth choices will conflict with the depth cues our brain gets from the overall image. For example, the size of characters in a shot tells us who is closer to the camera, or the depth of field can tell us what is close to the camera. Deficits in the depth process are most easily identified in stereo. However, it does take some practice see past the 2D image to understand what's wrong with the depth. Often it will just feel cheap, unnatural, or fake, leaving you trying to figure out what is wrong instead of being immersed in the story. A crane shot of a tree or foliage is often the easiest place to see depth issues. In 2D, a crane shot reveals which branch is closer just by its motion relative to the other branches. In 3D, the depth choices have to accurately compliment this motion. This can require hundreds of layers of depth. Another place to look for depth issues is at the edge of characters. Often the edge of a character will appear to roll-back unnaturally. This is due to inaccurately modeled or poorly matched-animated geometry. Acute stereo vision definitely helps to identify these issues, beyond seeing that something "just looks wrong". A practiced viewing technique can also help to "pop" these issues. When viewing a 3D shot, close one eye and then open it again. The scene will "pop" from 2D to 3D, and there may be areas that you'd expect to pop forward that don't, or vice versa.

When the matte process is short-changed it is fairly easy to see. If it's bad enough, poor matte work can make a live-action shot look like a bad composite shot! Flyaway hair may be completely removed. Hair may fall-back to the background. Holes in trees, fences, or other elements may appear to pull the background forward. Even transparencies may be completely ignored. If a transparency such as motion blur is not properly treated the motion blur area will appear to fall-back to the background or the background will appear to stick to the foreground character. This can create a distorting or watery effect. The same phenomena can happen with shallow depth of field. Another symptom of poor matte work is when moving characters appear to strobe. This is usually due to a hard-edged matte.

Deficits in the paint process usually mean the process has been bypassed or performed to a low quality. Foreground elements may appear stretched, streaked, dead/static, or watery. They may have artifacts, they may strobe, or transparencies such as smoke may appear flat. In an automated process, pixel-repeat or pixel-stretch is often used to fill in missing background information. This means the last pixel on the edge of a foreground is repeated until it fills the missing background. This will often look like an artifact or a blurry or messy edge. If you see a "dead" or static shot it may mean that a single matte painting was used to paint the background and the natural motion of the background was sacrificed (like windblown bushes or film grain). Elements that strobe are likely missing diffused edges that have been painted out, rather than properly treated for transparency. When transparencies such as smoke appear flat, it is often because the paint process has been short-changed.

Artistic Critique

The most subjective part of critiquing 3D is, of course, the artistic side. There are many schools of thought and contradictory rules. Some artistic approaches seek to ensure that the focal element resides at screen level. Others insist infinity should be the controlling force and should always reside at 2.5" of separation on the screen. Others try to find a middle ground that ensures a smooth transition of the focal element from shot to shot, while not massively pushing the scene past infinity. All of these approaches can yield good results in the right hands. Artistically, it becomes important for the director to know what they wish to achieve with 3D.

Let's look at genre as a potential starting point to artistically critique a 3D film. A goal, in 2D, for an action/sci-fi film, such as Avatar, is to create a realistic alien world. To do so the world must make logical sense and we must be immersed. For Avatar, since the film is also about feeling compassion for the aliens, the alien world is supposed to be awe-inspiring and worth fighting for. These goals for the 2D action/sci-fi film match nicely with respectable 3D goals. The 3D for an action/sci-fi should logically make sense, be immersive, and be awe-inspiring. All of these 3D goals underline the same story goals of the genre. Conversely, a goal for a slasher is to make the audience jump. So, the 3D version should be aggressive. Instead of an immersive look, the 3D might be used to suddenly encroach on the audiences' space to make them jump. In a psychological thriller, the audience should experience a visceral connection with the main character. Just like composition, angle, lighting, and movement is used as a visual metaphor that helps the audience understand the character's mental state or predicament, 3D can be used to the same end. In a drama, using 3D to portray the world realistically may be the most objective. In an animation or comic book movie, where interesting visual style is a staple of the genre, stylized 3D is probably a good artistic goal.

Understanding some 3D-specific phenomena may also help us critique the artistic side of a 3D film. "Miniaturization" is one of these. Often it makes the scene feel small, like the diorama you made in grade-school. Ironically, the more overall depth the scene has, the more miniaturized it will appear. This is the case because, in real life, we can only perceive depth within a relatively small range. In other words, we perceive distant objects as "flat", or lacking dimensional shape. For example, we can't perceive (in stereo terms) the separation between buildings in a skyline, or the shape of the moon. Another way to conceptualize this effect is by thinking of a sea battle between battleships. The battleships will appear miniaturized when appearing on the audience side of the screen. Our brain tells us that if the battleship is in the theater with me, it can't be that big. It must be a model because it has to fit inside the room.

Conversely, "flattening" may occur when there is not enough depth in a scene or object. If the audience doesn't perceive depth when they expect to, such as with a close-up of a person's face, the person will appear "flat." Or if a shot of an expansive environment, like a vista shot, plays close to the audience, it may appear overly flat. As an object gets closer to us, our brain expects to see more shape. As an object plays farther from us, our brain expects to see less depth in the object. So, if the vista shot plays on the audience side of the screen, it looks like a painting on a wall. But if that same vista shot plays at infinity, it will appear expansive and huge.

Depth continuity is also an area that should be touched on a bit more. The primary goal of depth continuity is to ensure the comfort of the audience. If your eyes don't feel tired after viewing the entire movie, it probably had good depth continuity. A secondary goal of depth continuity is to present a consistent world. As a basic approach, characters in various framings (close-up, medium, etc.) should have roughly the same amount of internal volume or shape from shot to shot. For example, if one character appears flatter than the rest during a dialogue scene, this inconsistency may pull the audience out of the story. This, of course, is subject to the artistic goals of the scene. In "Alice in Wonderland" the real world was supposed to play flatter than Underland because that mirrored Alice's boring existence in the world.

Go Critique 3D!

Hopefully with all this ammunition you can not only say if a 3D movie had good or bad 3D, but you can also explain where it went wrong or right. We should all hold this new industry to a high level of technical and artistic quality. It is possible to remediate and completely avoid the technical issues explored here. Ideally, the technical issues should be so minimal as to be invisible to the audience. Artistically, we should push for vision that compliments and expands the narrative of 3D movies.