Stereo Vision

Stereo Vision Tutorial

This page explains the principle of stereo vision and offers a simple tutorial to acquire the skill.

 

Introduction

Being able to visualize and experience structure in 3D is an essential skill if you are at all serious about understanding the molecules of molecular biology. This is not sufficiently realized in the field: many molecular biologists have never invested the effort it takes to learn the skill and thus will tell you that it is not actually necessary, and you can get by regardless. Of course you are talking to a biased population – unless you have experienced and worked with stereo images, you can't understand how much you are actually missing. But once you have acquired the skill, you'll regret not having been taught earlier. Speak to people who use stereo vision: seeing molecules in 3-D is like the difference between seeing a photograph of a place and actually being there. In 3D you can appreciate size, scale, distance, spatial relations all at a single glance. I insist: you can't understand structure unless you experience it in 3D.

From 2D to 3D
Images we see on screen or on paper are two-dimensional projections of three-dimensional objects.

Even though hardware devices exist that help in the three-dimensional perception of computer graphics images, for the serious structural biologist there is really no alternative to being able to fuse stereo pair images by looking at them. VMD is an excellent tool to practice stereo vision and develop the skill. Stereo images consist of a left-eye and a right-eye view of the same object, with a slight rotation around the vertical axis (about 5 degrees). Your brain can accurately calculate depth from these two images, if they are presented to the right and left eye separately. This means you need to look at the two images and then fuse them into a single image - this happens when the left eye looks directly at the left image and the right eye at the right image.

In this tutorial, I teach you a method to learn stereo viewing. The method is pretty foolproof - I have taught this many years in my classes with virtually 100% success rates. But I can only teach you the method – learning must be done by you.

Some people find convergent (cross-eyed) stereo viewing easier to learn. I recommend the divergent (wall-eyed) viewing - not only because it is much more comfortable in my experience, but also because it is the default way in which stereo images in books and manuscripts are presented. The method explained below will only work for learning to view divergent stero pairs.

Physiology

In order to visually fuse stereo image pairs, you need to override a vision reflex that couples divergence and focussing. This is something that needs to be practiced for a while. Usually 5 to 10 minutes of practice twice daily for a week should be quite sufficient. It is not as hard as learning to ride a bicycle, but you need to practice regularly for some time, maybe 10 or 20 sessions of 3 to 5 minute over a period of a week or two. Once you have acquired the skill, it is really very comfortable and can be done effortlessly and for extended periods. You will enter a new world of molecular wonders !

Instructions

Task:

Here are step by step instructions of how to practice stereo-viewing with VMD.

• Load a small protein into VMD (1UBQ will work just fine) and display this as a simple backbone model.
  • VMD Main → Representations → Drawing Method: Tube
  • Increase the tube radius to 1.0; choose coloring by Index.

A simple molecular model
 
For instructions to generate this scene, see above. To practice stereo viewing you want simple, uncluttered scenes with a moderate amount of occlusion and just enough shading to assist the brain in interpreting the image.

• Set the stereo display to SideBySide:
  • VMD Main → Display → Stereo → SideBySide

SideBySide stereo view
 
The image for the left eye is on the left side. For instructions to generate this scene, see above. Your practice window in VMD should have approximately the same dimensions.

• Resize the horizontal distance of the viewing window by dragging its lower right-hand corner. This changes the separation of the two views. Two equivalent points on the protein should be slightly closer together on the screen as the pupils of your eyes are apart (the average interocular separation is about 6.5 cm). Don't just guess, measure the distance, and adjust your on-screen scene to better than two or three millimetres of the correct separation.

Resize for the correct disance ...
 
You should resize the window of your molecular viewer (VMD) until equivalent points are the correct distance apart, they should match as nearly as possible your interocular separation. The two circles indicate equivalent points of the image. They are 232 pixels apart, their physical separation on the screen depends on your monitor resolution. On my screen they are 5.2 cm apart, too close for most people.

• Touch your nose to the screen and look right at the two images. Make sure you see the right image with your right eye, the left with your left eye. Of course, since you are so close, the images will be blurred and out of focus. Nevertheless, you should see one solid, three dimensional shape in the centre, plus two peripheral images of the same view on the sides. You see three copies of the same scene, but only the fused, overlapping centre scene appears three-dimensional; the other two become less noticeable as you practice more, your brain simply begins editing them out.

• Resize the window slightly until the centre image overlaps and "fuses". Slowly rotating the protein with the mouse helps generate the impression of a 3D object floating before you. (With VMD you can use the mouse to give the molecule a slow spin and it will continue to rotate). You could also type rock y by 0.2 50 into the VMD command window.

• Spend some time with this. Don't worry that it is out of focus. Imagine that you are looking at something underwater with your eyes open. But make sure that you see one central fused image and it appears three-dimensional to you. Don't continue unless you can achieve this. Ask for help.

• Once you see the object in 3D, try to move your head backwards slowly, until the structure comes into focus by itself. Do not voluntarily try to focus, since this will induce your eyes to converge and you will lose the 3D effect. After a short while, you will probably lose the 3D effect anyway. Once you loose the 3D effect, pause, look somewhere else, relax, take a deep breath and start over.

Practice this procedure patiently, two times daily for some 3 to 5 minutes. Stop, when your head feels funny. Don't force yourself.

After time and with practice, it will become easier and easier to achieve the effect. Also you will become quite independent of the distance of equivalent points, thus you can increase the viewer window size and take advantage of the increased resolution.

It should take you about a week or ten days to master this, with regular training it will become very easy. And, the best thing is, you do not easily forget this skill. It is like riding a bicycle, equalizing pressure in your ears while scuba diving, or circular breathing to play the didgeridoo: once you teach your body what to do, it remembers.

Examples

Here are example scenes to practice stereo viewing. You could also refer to the BCH441 Stereo Vision Exam Questions.

Abstract Shapes

Three staggered circles
 
Side by side stereo image. These are three circles staggered from front (top) to back (bottom). There are no secondary visual depth-cues at all, such as perspective drawing, size changes or occlusions. Once you see the circles in stereo however, their spatial relationship should become vividly obvious.

A cube
 
Side by side stereo image. This is a simple cube, drawn in orthographic projection and sligthly rotated. Other than the pseudo-perspective angling of the lines going towrads the back, there are no secondary visual depth-cues, such as occlusions or shading. This makes the 2D representation ambiguous regarding which face is closer to the viewer. If you look at only one of the images, you should be able to mentally invert the cube back-to front. In fact, when you concentrate on seeing the cube in a particular way, it becomes impossible to maintain this interpretation for more than abou ten seconds or so: your visual system forces the switch and thus explores all of the ambiguous interpretations. Once you see the cube in stereo however, it becomes clear that the face further to the left is closer to you. Concentrating on the image in stereo enforces the correct interpretation, rather than causing a switch.

A flat cube
 
Not a stereo image! This is the left eye image from above in duplicate. If you look at this scene like a Side By Side stereo image, it appears flat, and the back-to-front switching phenomenon occurs.

Left and Right Image superimposed
 
This is a superposition of the left and right eye views of the cube above, demonstrating the subtle differences that the visual systems uses to construct the appearance of three-dimensionality.

2MCP.PDB

An antibody Fab and its hapten
 
Side by side stereo image. The two chains of the Fab fragment of the phosphocholine-binding antibody McPC603 are shown in a cartoon representation, the light chain is in yellow (VL and CL domain), the two domains of the heavy chain are shown in orange (VH and CH1 - The CH2 and CH3 domains are part of the Fc fragment, not shown here). The bound hapten phosphorylcholine is shown in red in a surface representation.

Fold architecture in an antibody Fab
 
Side by side stereo image. The two chains of the structure have been color-ramped according to their atom-index in the PDB file to highlight the overall fold of the four domains. These are so-called greek-key beta barrels: each domain comprises a two-sheet barrel with a hydrophobic core. The chain meanders first through half of the first sheet, crosses over the top of the domain, creates half of the other sheet, crosses back along the side, completes the first sheet, crosses under the domain and completes the second sheet. In stereo, it is easy to trace the chain architecture by eye.

The antigen binding site with bound phosphorylcholine
 
Side by side stereo image. A transparent solvent-accessible surface, lightly shaded by position has been laid over a licorice representation of the protein atoms. A ball-and-stick (CPK) representation was used for the hapten. Elements are color coded red for oxygen, silver for carbon, blue for nitrogen; phosphorous is green. In this representation we can discern element types and bonding topology of the structure, as well as the shape of the binding site that the protein conformation has generated. A deep binding pocket accommodates the choline end of the hapten, the phosphate group is exposed to solvent. All molecular features of the hapten are discretely used in binding: the two oxygen atoms of a negatively charged aspartic acid can be seen at the bottom of the binding pocket, offsetting the positively charged trimethylammonium group of choline. Conversely, a positively charged arginine sidechain at the top, right, forms a saltbridge with the neagtively charged phosphate group. A tyrosine above the arginine hydrogen bonds one of the phosphate oxygens, while the hydrophobic sides of the binding pocket are formed by a tryptophan sidechain (top) and a tyrosine sidechain (bottom).

Common problems

Some people get confused as to what they are supposed to see. What you see are two images of an object and you see them with each eye, i.e. in principle four images. The two central images (image L as seen with the left eye, image R as seen with the right eye) should overlap in the middle; these two images fuse in your visual system to create one 3D image. The two peripheral images still remain; since you don't concentrate on them, your visual system will edit them out of consciousness as you gain experience.

There are two situations which interfere with stereo vision. One such situation is if the images presented to your eye are of unequal size. This can happen if you are using glasses with significantly different correction for each eye - the lenses then have different magnifications. The other situation is if equivalent points of the images are vertically misaligned, i.e. one of the images is shifted up or down, or rotated. This can occur when your head is tilted relative to the image. Keep it straight.

Images that are difficult to see in 3D are also images that are rendered differently for the left- and right view: non-aligned jagged edges, differing shadows or highlights disturb the stereo effect. To prevent this, try to apply visual effects judiciously. VMD is quite good with it's rendering but e.g. specular highlights in RasMol are quite inconsistent between left and right. (If you are a programmer, remember to write your code to move the camera location, don't rotate the object because that will incorrectly change shadowing.)

Of course, if you are learning Side By Side viewing, the so-called cross-eyed images won't work for you. They look like the real thing, but the left-eye view is on the right-hand side and vice versa. You can tell that they are wrong when you achieve the image fusion but the 3-D efect seems to be all wrong. This is because the image becomes inverted in depth: near points appear far away and vice versa. If you are looking at a simple line drawing, you can't tell that this is happening. However if there are any secondary depth cues in the image, such as occlusions, shadows, highlights etc, they are in the wrong places, won't work to enhance the depth effect and just generally look weird. Once you are comfortable with stereo viewing, you could try this deliberately by selecting cross-eyed display in VMD. It's very strange.

Further reading and resources

Stereopsis - Wikipedia article on why we see in three dimesions

Stereoscopy - Wikipedia article on the techniques for stereo images

Westheimer (2011) Three-dimensional displays and stereo vision. Proc Biol Sci 278:2241-8. (pmid: 21490023)

[ PubMed ] [ DOI ] Procedures for three-dimensional image reconstruction that are based on the optical and neural apparatus of human stereoscopic vision have to be designed to work in conjunction with it. The principal methods of implementing stereo displays are described. Properties of the human visual system are outlined as they relate to depth discrimination capabilities and achieving optimal performance in stereo tasks. The concept of depth rendition is introduced to define the change in the parameters of three-dimensional configurations for cases in which the physical disposition of the stereo camera with respect to the viewed object differs from that of the observer's eyes.