The Stereoscopic User Book
Transcription
The Stereoscopic User Book
THE STEREOSCOPIC USER BOOK How to make good 3D stereo pictures with a CineMonitorHD 3 DVi ew Alain Derobe All rights reserved © July 2009 Alain Derobe & Transvideo No part of this publication may be reproduced, stored or transmitted, in any form or by any means, in any countries, photocopying, including electronic imaging and any use on the net. A limited license for personal use is given to the user(s) of Transivedo equipment under his (their) responsibility. 2 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo Equipment that basically need the help of Cinemonitor HD 3DView display. B.Sp (BeamSplitter) setups : Two-camera rigs, set at right-angle with a beam-splitting mirror. S.b.S (Side by Side) setups: All stereo rigs with two side-by-side cameras. O.S (Other Systems) setups: All other two-camera setups, including setups with independent cameras on different tripods, sometimes far from each other as well as dual lenses or specific optical systems. Note : Professional stereo acquisition involves anyway a pixel-to–pixel synchronization with both cameras "genlocked" to the same timecode clock, and HDSDI monitoring outputs. OVERVIEW Stereoscopy is based on a dual point of view recording. To be able to see the deepness on a display requires wearing specific eyeglasses. (Only numerous point of view systems give the ability not to wear eyeglasses. The viewing corresponding range of depth is very compressed). The CineMonitorHD3DView have been created to clear the mind from any numeric datas or calculations, and shows 3D adjustment by direct comparative visualization. It gives a better way to resolve complex interaction in just a glimpse. The chosen way, Including viewer’s eyes specificities, didn’t prevent from a pixel to pixel accuracy, but better take in consideration the distinctive features of human sight, witch is never forget it, “the main goal“. As interactions between eye’s physiology and visual psychology are not yet well known enough to be easily set in data or equation, visual adjustment on a screen is at the moment the better way to set good 3D Then, we better use “off set“ expressing, speaking about distance of homologous items in left and right pictures, (measured on the screen in pixel unit), than “parallax” that involves boring angles calculations. In addition, we never know if “parallax“ term is used speaking about taking pictures or as the result in screening, which makes confusion. Therefore, we will see through a glossary at the end of this user’s book, that those two fields have to be completely dissociated in 3D stereo strategy. Convention: The terms used for shooting are underlined. The terms concerning projection are in italics. For instance, we will better use terms such as “angulation“ (in % of picture width), interaxial (in cm or inches) and convergence plane (in meter or feet) as a result of the combination of the two formers. In this “stereoscopic user book“, a step to step process have been designed to drive 3D professional consultant as well as 3D beginner, to set 3D cameras adjustment in a right logical way. “Natural depth method“ additional book drives the reader with virtual situations, in a justified and a structured strategy on the path of the fundaments of the stereoscopic system. Both will give to the reader, answers to most of 3D problems. The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 3 4 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo MANUAL SUMMARY CineMonitorHD3DView Allows Specific Stereoscopic Adjustments ............................... 7 A. 1. PRELIMINARY CAMERA GEOMETRICAL ADJUSTMENTS. (Basic geometrical disparity). 9 1.1 TILT - IMAGE HEIGHT ADJUSTMENTS: ........................................................................ 10 1.2 ROLL - RELATIVE HORIZONTAL ADJUSTMENT........................................................... 11 1.3 ZERO ANGULATION CALIBRATION:.............................................................................. 13 1.3.1 FRONT END CHECKING. ......................................................................................... 15 1.3.2 Having the two lenses not exactly at the same height may have drastic consequence on the image height adjustment...................................................................... 15 1.3.3 Mainly with macro stereoscopy, note that if one lens is in front of the other, the background will be correctly superimposed but not the foreground. .................................. 15 2. SHOT BY SHOT GEOMETRY ADJUSTMENTS AND SPECIFIC ADJUSTMENTS ACCORDING TO FOCAL LENGTHS. ............................................................................................ 17 3. CHECKING ANGULATION FOR EACH FOCAL LENGTH. ................................................... 19 4. CONVERGENCE PLANE POSITIONING AND MASTERING. .............................................. 24 5. USING THE GRID - DISTANT PLANES POSITIONING - COLORED BORDERS EQUAL TO GRID PITCH - WIDER OR NARROWER COLORED BORDERS - HOW TO CALIBRATE EFFECTS COMING OUT OF THE SCREEN .................................................................................. 26 6. GRID USER CALIBRATION. GRID SPACING ADJUSTMENT ACCORDING TO FINAL THEATER SCREEN WIDTH............................................................................................................ 27 7. DETECTING SPECIFIC STEREO DISCREPANCIES. ........................................................... 28 B. CineMonitorHD3DView helps to achieve two-camera photometric matching. ............ 29 1. USING DUAL HISTOGRAMS. ............................................................................................... 30 2. USING DUAL WAVEFORM MONITORING ........................................................................... 30 3. USING DUAL VECTORSCOPE. ............................................................................................ 31 4. DETECTING FLARE DISSYMMETRIES AND BRINGING FORWARD UNWANTED OR UNMATCHED GLARE. ................................................................................................................... 31 5. CHROMA FLICKER OR SWITCHER. .................................................................................... 31 C. OVERVIEW OF A STRATEGY: “NATURAL DEPTH™“ METHOD. ......................................... 32 1. Rules for shooting and displaying depth. ............................................................................. 32 2. Working method: Step to step strategy................................................................................ 33 3. For which screen width can adjustments be performed when the final projection size is unknown?........................................................................................................................................ 34 4. SCREEN WIDTH AND MATCHING BACKGROUND OFFSETS........................................... 35 5. Two important bounce effects are of major importance to understand what is actually different on large screens............................................................................................................... 36 6. HOW TO BLOW OUT THE SCREEN BUT NOT THE VISION. ............................................. 38 D. GLOSSARY ............................................................................................................................... 42 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 5 Recommended setups for stereo acquisition. The CineMonitorHD3DView does not involve any specific test chart nor particular equipment. It is thus possible to adjust the equipment on location, during shooting, with the actual sets. Nevertheless, apart from the usual camera assistant equipment, used for conventional adjustments, the following additional items could be useful: - A black flag with a small white cross will be set up on a stand in the frame, to simulate a convergence point, when a character is not in his position or when no element stands out in this specific area. - On the other side, a double cross matching exactly the interaxial distance of a side-by-side rigs, will help adjustment when pointing indoor short distance for parralelism control. - A light-toned, over 5' (1.5 m) long rod will help to adjust relative horizontal level. - When complex special effects are planed, an angle to level measuring tool and a laser beam pointer instrument are very helpfull. - Color test charts to help respective color rendering fine adjustments on the two cameras. Preliminary warning: -When looking to the monitor with “red / blue-green anaglyphs” without wearing the glasses, keep in mind that a light-toned picture element with a red border on its left, means that this item will be shown to the viewer behind the screen plane. On the contrary, if the element is dark, a cyan border will show on the left for the backgrounds. -It is very important to get used to this discrimination to obtain an immediate reading of the L/R colored separation given out by the monitor. -Of course, a light-toned item with a red border on its right will show a coming out of the screen effect, just like a dark item must have a cyan border at its right if it is located in front of the screen. -With monochrome magenta/green function and appropriate glasses, ones can see as well deepness or better, but without color rendering. When light bright primary tones are in the field, this system will give less colored artifacts. -Yellow to blue anaglyph system could be available on special order. All those anaglyphic systems are very useful to inspect the picture, without wearing glasses, and shows perfectly what belong to right or left picture. Active glasses system, gives a nicer deep visual effect for fun, but didn’t give any identification about left or right pictures adjustments. This function is very useful to simulate a residual “ghost effect“ on the screen when lightly areas are on a very dark background. 6 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo A. CineMonitorHD3DView allows Specific Stereoscopic Adjustments The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 7 8 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 1. PRELIMINARY CAMERA GEOMETRICAL ADJUSTMENTS. (Basic geometrical disparity). To achieve these adjustments, it is recommended to begin with the focal length you plan to use the most frequently. If not, choose the normal focal length closest to the horizontal width of the sensor(s). It is the focal length that will be closer to an orthoscopic vision better for stereo. Adjustments for other focal lengths will thus be distributed on both sides of the normal focal length. For instance with 2/3 inches chip ( 9.5mm horizontal HD size), a 9mm or 10 mm focal lens will be considered as normal focal lens. (Diagonal is better a reference to still picture work). After setting up the equipment and connecting the cables according to the operating instructions in the “Operator Manual“, set the monitor to “3D function” (for instance, red and cyan). Check that the two lenses are positioned at the same level by inspecting from the front of the rig. If not, the frame heights will never match for both distant and close-up shots. Note that with general use from 3' (1m) to infinity, a 1 mm lens height difference or more can remain unnoticed, when it gives some troubles with macro or very big close up. With the beam-splitting mirror rigs (B.Sp), overlap the two lenses horizontally as well as vertically, by looking through the mirror. Mark that starting position on the spacing scale as the null interaxial distance. The cameras will be perfectly adjusted when the dual picture on the monitor will not show any colored borders On most rigs, the left camera will be the horizontal one. Choose it as reference camera. (Note that on stereo HD Steadycam™ rigs, it will often be the reverse.) With Side-by-Side rigs, (S.b.S ) adjustments are generally made by pointing at a very distant scenery. If not, then use carefully the appropriate double cross target. Levels of the lenses have to be controlled physically with a rod. The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 9 1.1 TILT - IMAGE HEIGHT ADJUSTMENTS: The camera rig has build in tilt adjustments. (If not, use spacers at the back of one of the cameras to obtain a proper adjustment). Point a target and check there are no colored borders above small horizontal elements. Choose relevant items, preferably at the same distance from left and right borders to avoid any confusion with the Roll adjustment. (If you think the two lenses are perhaps not strictly on the same level, aim better at background items.) When a light-toned item is upper lined with a red border, or if a dark item is upper lined by a cyan border - like dark scenery on a background sky, lift up slightly the front of the left camera (red picture). Use the display's 2x enlarging feature (F4) centered between right and left sides (using blue thumbwheel) to fine adjust with a one-pixel accuracy. (Fig.1) B.Sp. (Beam Splitter system). The adjustment is very easy for rigs with a beam-splitter mirror system, if the two lenses have been previously perfectly overlapped by checking from the front. With a null axial distance, the two pictures should then be strictly identical on their entire surface. The smallest colored border shows there is a camera matching discrepancy. (Possible defects: When the beam-splitting mirror is not strictly flat, or when the two lenses do not produce perfectly overlapped pictures, tiny colored borders may appear only in a particular picture area. For instance a strain on the mirror can be implied or other better matching lenses may be found. If the colored borders are over a few pixels, we suggest recording a grid or a constant-pitch cross hatch to make systematic correction in post-production easier.) S.b.S. (Side-by-Side system). For outside shots, better find a distant landmark so that the two pictures can be overlapped perfectly. Otherwise, choose elements with horizontal lines, to compare them with their own extension. No need to say that with side-by-side cameras, any item that is not distant shows colored borders on the side. When tilt adjustments are very small, you may prefer making fine adjustment during postproduction. In this case write down the number of pixels for each focal length, to make later correction work easier. A 1 to 3/1000th of the frame height precision is acceptable when shooting feature films (one pixel accuracy is more professional work, but a 1% error for a very short time will not prevent the majority of the audience to merge fleeting images. (Note : Watch out for mechanical stress or play due to the weight of the cameras that can slightly modify the vertical adjustment during extreme up and down tilt movements. We suggest testing these positions during preparation and control an incidental baseplate loose). 10 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 1.2 ROLL - RELATIVE HORIZONTAL ADJUSTMENT. The camera rig has a build in horizontal roll adjustments. (If not, use spacers under the side of the camera adaptors to obtain a proper adjustment). Watch now alternately and in the same manner the upper right and left corners with the 2x enlargement. Upper or lower colored borders with different colors from one corner to the other reveal a picture roll between the two cameras, meaning a relative horizontal discrepancy. B.Sp. In the left corner of the screen, a red border above light-toned elements means that the left camera leans to the left. Lift therefore its left side. Do the opposite if the upper colored border is cyan. In the right corner, it would be the opposite. (Fig 2). Note : Remember, you can be confused when dark elements on a light-toned background show reverse colored borders. Aim at horizontal lines as much as possible. Slanted lines are misleading! Viewers with delicate visual systems, even on large screens, will tolerate a vertical mismatch of some 2 or 3 pixels in HD resulting in .02% to .03% of the picture height. However, trying to obtain a one-pixel precision would be better for long durations and to achieve professional quality. Correcting roll mismatch in postproduction requires re-sampling and can destroy image quality. Horizontal roll adjustment must then be done precisely as it is final and focal length independent. After roll adjustment have been done, it can lead to a minor tilt adjustment (§.a). Check if needed. S.b.S. For outside shots, finding a distant landmark (horizon, buildings) is the best. For interior shots, look for a horizontal architectural item perpendicular to the optical axes (i.e. furniture, railing) or set up a light-toned horizontal rod on a stand at a right angle with the axis of the shot and covering the entire frame width. Check as above for reverse colored borders from one side to the other. The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 11 Figure 1 – Vertical tilt mismatch Figure 2 – Clockwise left camera roll 12 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 1.3 ZERO ANGULATION CALIBRATION: The camera rig has build in angulation adjustment. (If not, turn the camera horizontally). First of all, any angulation adjustments on Beam Splitter Rigs have to be set with the two lenses in absolute superimposition through the mirror. With Side by side rigs or other systems, lenses axes have to be perfectly parallel. B.Sp. With beam-splitter mirror rigs and a null optical interaxial distance, pictures should be perfectly superimposed. If colored borders appear, correct the overlap with the angulation adjustment and reset the zero mark. (This mark can be slightly different according to the lenses.) This mark is mainly a reference one for user information, because the CineMonitorHD3DView accounts for all accidental or repeated drifting. Nevertheless, zero angulation will be useful during parallel shooting, either for an adjustment aimed only at very large screens, or for making some special effects easier (i.e. tracking). S.b.S. When a very distant background is available (more than a thousand times the interaxial distance), adjust the angulation until the overlap has no colored side borders. (i.e. with an 6" (15cm) interaxial distance, choose a background more than 500'(150m) away). For instance, at 1000ft (300m) a one pixel residual offset will remain. If the item is not far enough, then try to appreciate the interaxial distance directly with the subject, if it has known features. For instance, on a relatively close building, shift the two pictures by comparison with the width of a brick. That can be accurate enough. For inside shots, use a double cross with exactly the same offset distance as the interaxial distance. A zero angulation marking can thus be corrected or defined. Bear in mind that the angulation adjustment as well as the resulting convergence are generally tweaked in post-production and then are not critical settings. O.S. (Other systems). Two cameras positioned separately can also be calibrated and lined up. (Of course, they must be re-calibrated for each setup). -- For very distant landscapes: Mountains, etc. After having carefully checked the double tilt and roll position of the cameras, proceed in the same way as for S.b.S. concerning the three main adjustments by aiming distant items. It is possible to adjust very distant cameras with a long BNC connexion. Just frame the first one and then adjust visually the level of the other one with a fine tilt and roll adjustment knobs. -- A specific case: Hyper Stereo on shallow subjects : For instance, for tight close-up shots or to increase the depth of a “bas-relief” on a monument, with a very large interaxial distance. To obtain Hyper Stereo, a very strong angulation (several degrees) must be applied, bringing inverted trapeze distortion (keystone cross effect) on the two pictures. Image height (like in "A") and angulation (like in "C") adjustments must be performed as close as possible to the center of the frame to overcome the trapeze effect. You must next evaluate the (as describe in "B") roll error by choosing only elements half way up the frame or by judging by sight the same symmetrical colored border values at the right and left of the frame. Especially for small depth values, the unescapable trapeze distortion will be partly corrected in post-production. Note that for an angulation usually set for a normal stereo effect on characters (generally under 1°), trapeze distortion remains below the audience's visual tolerance, even on large screens. The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 13 Figure 3 - Zero angulation calibration 14 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 1.3.1 FRONT END CHECKING. With beam-splitter rigs, check from the front if the two lenses are correctly superimposed vertically when the two lenses are at null horizontal spacing adjustment. 1.3.2 Having the two lenses not exactly at the same height may have drastic consequence on the image height adjustment. A height mismatch between foreground and background occurs on the screen only if the height difference between lenses is a significant part of the closeness of the foreground. By instance, with a normal focal lens and a HD 1080 setting, a one centimeter height difference (vertical parallax error) with an adjustment to infinty, wll create 1 pixel mismatch for a local 17 meters foreground, and 10 pixels uneven level for what is at 1,77 meter camera distance. See the diagram showing an inability to get a mountain, a tree and close-up character, leveled in a same time when lenses are not leveled. When you place horizontally two or three wires or strings in the deepness of the field, CineMonitorHD3DView will help you, in showing colored underlines, to correct unnoticed lenses height mismatch. (Fig. 4) 1.3.3 Mainly with macro stereoscopy, note that if one lens is in front of the other, the background will be correctly superimposed but not the foreground. When you suspect that the two lenses are not exactly at same distance of the subject, place a large graduate ruler at 1meter distance from the normal focal lenses. If one picture is 1cm larger than the other on the CinemonitorHD 3Dview, this means that one lens is 1 cm behind the other one. The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 15 Figure 4 - Height difference 16 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 2. SHOT BY SHOT GEOMETRY ADJUSTMENTS AND SPECIFIC ADJUSTMENTS ACCORDING TO FOCAL LENGTHS. Note that if the rig is rigid enough, roll adjustments will remain unchanged whatever the focal length. You must of course keep track of the lenses, mark them with colored tape, red for left one and blue or green for the right one, and never switch them around from one camera to the other. It is indeed well known that axis offsets can occur depending on each specific lens. This can come from either an offset between the lens axis itself and its mount or more often from the offset between the sensor(s) and the camera mount. In this case the offset will regularly progress together with the increase of FOCAL length. Axis offsets coming from lens mounts appear when switching different similar lenses on a same camera. Even with the most professional equipment a slight axial mismatch does not guarantee a sufficiently precise adjustment to shoot in stereo. The adjustments made with the monitor offer the advantage of taking them into account. When the lenses have an available re-centering adjustment, such as the 17/112 Zeiss Digiprime zoom, the CineMonitorHD3DView helps evaluating the adjustment to be done. With other systems, a very delicate work in a specialized workshop can be planned. a)- Then, a vertical correction may have to be done when the focal length changes. It can be done on the spot and by sight as in paragraph § 2-1 a. It will remain always identical to itself with the same lens, and can be marked on the tilt adjustment knob (if so) to anticipate and spare time. As long as tilt corrections remain small, the shot by shot adjustments by sight will be almost immediate. When they are very small, some will choose to leave them to post-production, even if a few pixels have to be cropped.. b)- When the CineMonitorHD3DView monitor is used in stereo mode, the horizontal adjustment for each focal length is of course included in the adjustment by sight. When a zoom is used like a series of fixed lenses, these adjustments can be made on the fly. A dual remote control system will help. Fig 1-2B For Daredevil people, some small focal re-framing can be done during the shot if a double zoom motor is available. It must be extremely reliable because human vision does not tolerate unequal image sizes. To remain on the safe side, focal lengths should be below twice or three times the normal focal length. Nevertheless, when shooting distant scenes with long focal lenses and/or large zoom amplitude, such as sports or performances recording, sophisticated equipment with integrated "Motion Control" must be used, with all the corrections pre-recorded on the computer. The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 17 Figure 5 - Focal length mismatch Figure 6 - Correct focal length adjustment 18 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 3. CHECKING ANGULATION FOR EACH FOCAL LENGTH. With each new shot, to keep distant landscape always similarly behind the screen and to obtain the same depth view for the audience, the angulation adjustment must physically be performed, depending on focal lengths. Actually, to obtain the same offset on the screen, (6,5 cm max for the background) the applied angulation must decrease when the focal length increases. (Fig. 7). Figure 7 - Far Background adjustment The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 19 Please refer to the "Natural Depth" tutorial, and remember that, whatever the size of the screen is, items corresponding to background areas, have never to be offset over 6,5 cm (2,5 inches) on it. This absolute rule ensues the physiological impossibility of having diverging sight. With the CineMonitorHD 3DVIEW, the adjustments displayed on the monitor are the direct results and do not depend on the focal length used. (Fig 1-3 bis). With the user-defined grid pattern, a grid pitch is chosen generally during the whole continuity according to the size of the final screen. Then a constant “scenic box“ can be preserved (or breack out), depending on the option taken by the stereo consultant. Anyway, and whatever the lens used, offsets between left and right picture elements, that show farest background items have to be adjusted with the choosen grid pitch. To have a grid pattern instead of a simple crosshatch avoids moving the camera away to aim at a particular element, and also give the opportunity to inspect all distant far planes together with one look. Note : For those who want to work with parallel cameras, and who wish to simulate convergence, proceed in the following way: Use the "Shift" function of the CineMonitorHD3DView monitor, and shift the left picture (red) of the equivalent of one step of the grid towards the left side of the picture (this means the same number of pixels than the chosen grid pitch). This "virtual angulation" will directly display on the screen the result of the convergence, which will be reconstructed with an appropriate offset in post-production. The convergence plane will thus be simulated (stereo window) and the same depth than with a real angulation will be observed with 3D stereo display function. 20 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo Figure 8 - Angulation Adjustment with various focal length The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 21 22 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 23 Figure 9 - Angulation, Separation and Convergence Plane 24 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 4. CONVERGENCE PLANE POSITIONING AND MASTERING. An Angulation ( or a replacement "offset" simulation) having been previously positioned, as mentioned above, the right way to change the convergence plane position. will be in modifying only interaxial distance. (Fig. 9). Perform a systematic adjustment of far backgrounds in setting correct angulation “α”, then adjust convergence plane “C” by setting camera interaxial distance “ε” Three type of elements will then be observed in the picture : - All elements with light-toned parts, which show red borders on their left, will be the distant planes. - All elements whose light-toned parts show red borders at their right will be coming out of the screen. (The elements coming out of the screen and that are cut off by the edge of the frame will quickly be identified as disturbing). - Then, all the areas with no colored borders are basically on the convergence plane. (Normally, the convergence plane is supposed to be displayed on the screen plane, unless further horizontal offset adjustment during post-production.) Any change of interaxial distance will rearrange these areas with an obvious display of colored borders, so the balance between them will be done at a glimpse of an eye. Warning: Viewing with anaglyph (red and cyan) as well as active glasses (LCSs Liquid Crystal Shutters) will display a pleasant 3D effect but with offsets and angulation adjustment not suited for a large screen. The grid pattern is the only standard that allows calibration for large screens. The convergence plane resulting from the adjusted interaxial distance will be positioned with the following rules: Strictly speaking, no element coming out of the screen should normally ever be cut off by the edge of the frame. This means that the convergence plane should always be in front of the nearest element. This rule can particularly be annoying, during dialogue scenes with a character cut off by an over-the–shoulder shot! Usually you may cheat a little, but not up to the point where the audience becomes aware that you do. Recommendations on this “stereo window error“ are found in the paragraph about the "Natural Depth" method and in detail in the corresponding book. The front parts of the fence come out of the screen. The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 25 5. USING THE GRID - DISTANT PLANES POSITIONING - COLORED BORDERS EQUAL TO GRID PITCH - WIDER OR NARROWER COLORED BORDERS - HOW TO CALIBRATE EFFECTS COMING OUT OF THE SCREEN All evaluations, all reasoning as well as precise calculations on the position of stereo picture elements in the theater, become miraculously evident when an interocular distance (IOD – Inter Ocular Distance) of 2.5" (6.5cm) is used as a standard unit. For instance, the 2.5" (6.5 cm) IOD calibration will fit hundred times in a 25' (6.5m) theater screen and corresponds to a x100 pitch displayed on the grid. (See figure in the central page) So even if you don’t know the size of final screen, you have to choose one width to make a consistent continuity. Even if you intend to cheat, keep the grid pitch at exactly the ratio between IOD and theater screen size, and prefer, if necessary, to calibrate visually beyond the grid pitch. For instance, some far plane offsets beyond the distant elements calibration requiring little eye divergence from the audience can momentarily be tolerated. 8,10 cm or (3, 4 inches) measured on the screen. Regarding this, the audience beyond the “Depth bracket” will see confined elements such as a far light in night background. Then the “Scenic box“ will locally overpass “Depth bracket” limits. The grid pitch representing exactly the IOD will also be very useful to appreciate the other depths and specifically to quantify effects coming out of the screen. Then, It has to be used as a general reference. In this case, not only the distant elements will easily be calibrated, but (when a reversed offset) the elements coming out of the screen showing the size of one, two or three column intervals, will be respectively at half, one third or one fourth of the distance between the eye and the theater screen. It is the only way to define how much is the extension of the outcoming of the screen, while trying to measure them in feet (or meters) would not make sense. Can you imagine an easier system? Thus, you can learn to "read" rapidly directly on the TransvideoHD 3D display the position of each depth element. Isn’t it every stereo consultant dream? Consequently, every one scale division reversed offset (the red border of light element being on the right side) will point out the elements coming out of the screen at halfway between eye and screen. This means ½ OSD (ocular screen distance) : see diagram "E". The bottom of the table on the graphic, simply shows that one, two, three or five crossed offset intervals show elements coming out of the screen respectively at one third, one fourth or one sixth of the eye-to-screen distance. ( see "F, G, H" diagrams for : 1/3 OSD, ¼ OSD or 1/6 OSD ). Intermediate values will be properly appreciated at a glance. For instance, an element in the rear with a colored border half the size of the grid's pitch value will be seen halfway between farest plane and screen plane for the whole audience whatever is the position of the viewer. (see "C" diagram). Is’nt it easy to use ! 26 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 6. GRID USER CALIBRATION. GRID SPACING ADJUSTMENT ACCORDING TO FINAL THEATER SCREEN WIDTH That 3D stereo pictures must be "calibrated" according to a given screen size, comes as a surprise to many ! What can then be done to account for different theaters ? Fortunately, depth can be readjusted according to screen size with little trouble and minor losses by a full frame sideways offset. The “Natural Depth™“ method suggests two typical screen sizes: 21' (6.5 m), the most convenient for any purpose use and as a shooting easier reference for beginners, and 30/33' (9/10 m), better for feature or professional master delivery. For instance, in HD and for a 30' (9 m) screen (140 times the interocular distance: IOD), a 140 lines grid (14 pixels in HD) will be used to calibrate immediately most of the far planes at their safe value (See figure in the central page) Note that for larger screens a grid with a narrower pitch would not be easily readable ! Therefore, for screens 42' or 85' wide (13 m or 26 m), you can refer to 1/2 or 1/4 grid interval while keeping a pitch of one hundredth. Some will prefer using the picture offset, or the user crosshatch, both being equally adjustable pixel by pixel. In this case the stereo 3D display on the monitor is degraded, there is no more help for adjusting the interaxial distance and the information about the elements coming out of the screen is lo Nevertheless, one can think that with this giant size, the convergence plane obeys different laws and the interaxial distance will be chosen according to other criteria. As window error decrease with large screens we recommend to work as if for a 21’ (6,5 m) or 33 feet (10 m) width screen and recalibrate if necessary for larger ones. -For resolutions other than HD 1920, apply a rule of three, depending to the display and the pixel numbers (for instance with 720p). -In an other hand, for smaller screens than 6,50 m (21 ft), the far planes offset limit is no longer proportional to the interocular distance. Indeed, there would be no sense in offsetting close to a half of the whole picture on a 7" monitor!!! The offset of the far planes on the screen, then have to be asymptotic to a fixed limit of around 2 to 2 ½ %. This represents a maximum limit of about 40/50 pixels in HD. See the graphs corresponding to screen widths, from the "Natural depth" method. (Numbers can be a little corrected, depending of the distance from the display to the viewer. When watching still pictures this distance is usually bigger. Then the “one thirty’s rule“ that gives 3% of the distance to calibrate the offsets, cannot be applied to movies work.) The cause lies in an inescapable "correlation" (called synchrony) between the viewer's crystalline lens focusing on the screen (accommodation) and his two ocular axes attempting to converge on a point at another distance. This physiological tyranny will particularly prevent displaying far planes beyond a very small fraction of the width of small screens, because decorrellation possibility is very restricted. Consequently, with these small display widths, the elements coming out of the screen, far from being proportional to their size, will be extremely reduced and flattened. You can now understand better why even when you display stereo on the monitor for safety, or to check for some discrepancies, YOU HAVE NOT TO CALIBRATE THE DEPTH AMPLITUDE WITH THE STEREO EFFECT YOU CAN SEE ON A DISPLAY SMALLER THAN THE FINAL ONE. ANYHOW YOU MUST COMPLY WITH THE GRID INDICATION TO GET A FITTED WORK. The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 27 7. DETECTING SPECIFIC STEREO DISCREPANCIES. In stereo anaglyphic color mode, (with or without the eyeglasses), or in stereo magenta/green monochrome mode, it will be easy to detect an isolated colored element or an element far away from its counterpart in a complementary color. These solitary elements that strike the eye are called "orphans". It is often a leaf, a gaffer tape or a connecting wire hanging close to the camera and left unnoticed. Superimposing on the monitor also points out a French flag in the field of only one of the cameras. Actually, using the main viewfinder can make the camera operator forget what is happening on the other viewfinder, often neglected because of its upside-down picture on B. Sp. rigs or a lack of room to place the viewfinder on S.b.S ones. All set element partially in the field at the frame edge on one of the cameras and not on the other will then be easily eliminated. In the same way, "flares" and asymmetrical reflections, which are the greatest curse of stereo picture pairs, will be easily eradicated by observing two superimposed pictures on the same display. A particularly sneaky defect with a regular pattern can appear: when for instance the offset between the two images is the same as the pitch of a grating or any similar frequency like a fence or an iron gate. In this case the superimposing attempt made by human vision becomes random and hesitates between several neighboring patterns since they are similar. The perceived effect is extremely unpleasant. The grid will then be a valuable help to detect the dangerous areas. 28 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo B. CineMonitorHD3DView helps to achieve two-camera photometric matching. The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 29 1. USING DUAL HISTOGRAMS. As well as a waveform allows to appreciate (in an analog way) the light amplitude displayed vertically, the iris differential adjustment is made easier by the huge lateral offset displayed by a double histogram. Actually, if a specific area with the same brighness on each picture can be spotted, this area will clearly undergo a digital category change over the histogram. Mostly its central significant part, like its hight lights right part, will move rapidly sideways at the smallest exposure change. Comparative lighting adjustment can therefore be executed with very high precision, way beyond what just matching the two pictures requires. Some mismatch on the two envelope between homologous areas will show up discrepancies in specific high or low lights. That can reveal a sneaky one-eyed glare or reflexion. 2. USING DUAL WAVEFORM MONITORING Double “waveform“ will remain useful to locate within the picture width if a specific reflexion is clearly more intense on one of the cameras. In general, it is advised to be concerned each time a waveform as well as a histogram envelope difference is noticed. For instance, if a different gamma or a white compression circuit does not match perfectly the other camera, then you will see different "waveform" or “histogram” envelope. 30 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 3. USING DUAL VECTORSCOPE. Masking in turn each camera can show up a color shift. Responses on color charts can also be compared. White and black balance can also be checked. 4. DETECTING FLARE DISSYMMETRIES AND BRINGING FORWARD UNWANTED OR UNMATCHED GLARE. A flare or an unwanted reflection is visible because it pulls up the black level. It can be detected either if the left side of the double histogram is not overlapping, or if the bottom of the double waveform is lifted up. 5. CHROMA FLICKER OR SWITCHER. All general or local synchronization defects generate a particular flicker on the monitor. This unpleasant picture instability is useful to warn the user that a temporal offset is present between the two pictures and that the audience can be disturbed, for instance by car wheels on a pan shot. The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 31 C. OVERVIEW OF A STRATEGY: “NATURAL DEPTH™“ METHOD. 1. Rules for shooting and displaying depth. The "Natural Depth™" method aims at converting the real depth of the subject to obtain a limited depth called "Scenic Box" through the theater’s screen. It is well known that the Scenic Box shown in projection is different for each viewer seat. The paradox is that after a few moments of accustomation, it will nevertheless appear as a fair likeness of reality to every viewer at any seat at whatever position ! This is a visual adaptation phenomenon and that is fortunate, otherwise who would buy tickets for one-seat theaters! - In addition, the Scenic Box is practically never identical to natural layout. It obeys its own rules, witch are very different ones from natural seeing the surrond. Therefore, the comparison between the goal - a satisfying “Scenic Box“, and the real depth of the set will lead us to the correct camera adjustments of the appropriate rig, whatever type it is. Also, very different extensions of “Scenic Boxes“ cannot be successively displayed on the screen whatever reality is represented, but a visual presentation continuity must rather be preserved. In all circumstances, try to fill as better as possible the volume of the Scenic Box, without minding anymore the real subject amplitude. The maximum “Bracket of Depth“ cannot be used all along the continuity because of eyestrain and also because that did not give anymore adjusting possibilities to post production. So effective “Scenic Box“ will be generally less extended than maximum “Bracket of depth“. Compressing or expanding reality seems then of little importance by comparison with managing on the screen a flow of several Scenic Boxes that must follow their own laws. These laws aim at satisfying the audience's vision and are an absolute priority, whatever the subject or how it is set up. 32 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 2. Working method: Step to step strategy. Look out for elements the farther away behind the subject, containing details that can be analyzed. They will be fare backgrounds, and their left and right images must be shifted on the display to be seen as much behind the screen as possible, preferably at the rear end of the Scenic Box. (That means a 6.5 cm offset on large screen whatever its size. That offset have not to be much overpassed). By applying an angulation between the two cameras axis, the resulting offset between these far planes will be correctly adjusted on the screen. II should match the grid pitch on the display. (Remember this pich has been chosen according to the final projection width size.) This offset can be slightly readjusted in post-production by a global offsetting of the entire image to make shots even. This small correction does not change much the relation between background and foreground planes. Applying different optical interaxial distances to the same securely fixed angulation will consequently push back more or less the optical axis crossing. This is how the convergence plane will be adjusted forward or backwards, following the well known triangulation laws. Changing convergence plane position during a scene will then have to be a modification of interaxial distance, not of angulation. Whether axial distance modification is made during the shot, or if it concerns successive shots, the audience will not notice any variation. With moderate interaxial movements, various depth elements will be neither flattened nor distorted and will quietly stay in position for the audience. For the same screen width and with far planes at infinity, there is only one correct angulation per field width, hence per lens. For each screen width, the number of grid intervals must fit to the number of times the interocular distance holds inside the screen width. Whenever a close obstacle, such as the wall of an indoor shooting, limits the distant background of the picture, the angulation must be increased until the offset indicated by the grid is obtained again. Of course, this interior wall is now a “restricted background“, but it haves to be considered as well as the new far plane in the scenic box. On the other hand, those who will choose against common sense to modify the angulation while a character is getting closer, to keep the convergence plane on him, will necessarily greatly increase the far plane offset on the screen. Beyond twice the intraocular distance on a medium size screen, a large part of the audience will not be able to merge anymore left and right pictures, and will see them double (diplopia) while, with the rest of the audience, eyes will hurt or constant headaches will appear. This unlookable background is much worse than a possible miniaturing effect or gigantism effect. How to be protected from these abnormal size effects will be explained in detail with “Natural Depth™“ method. Benefits from using a stereo 3D monitor. Of course, when screens of any size, lenses with different fields and various distances are concerned, adjustments could eventually be meticulously calculated every time by geometry drawing or with tables, software and formulae, but this is not the right way for a rapid setup on the stage. With this way of calculation, how appreciation or cheat can be applied depending on the nature of the image, when human comparative feeling about significant areas in the picture is a basement of 3D stereoscopic language ? Formerly, when adjustments were executed blindfolded, mainly with film stock shootings, there was no other possibility than these sluggish calculations. Now the revolutionary CineMonitorHD3DView, allows precise real time adjustments, without any calculation or tables. It also gives the possibility to modulate what the stereo operator can (and must) appreciate regarding picture composition. The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 33 3. For which screen width can adjustments be performed when the final projection size is unknown? The use of a standard based on 21'(6.5m) screen for shooting, in addition from using a very practical 1/100th (20 pixel) grid, corresponds directly with an average width of projection screens used for screening or post production control. Professionnal feature film makers will like better to work straight to very large screens with a standard like a 10 m (33’) one, to get less trapezoid distorsion. Then they will better use a 14 pixels grid (14 pixels are 140 times in a 1920 pixels picture width). When using the 21’ (6,5m) width screen you have to strictly respect the stereo window rules. It is an additional mean to use it as a reference. (We remind you that “stereo window error“ is the visual contradiction caused by an image element seen in front of the screen being caught by the edge of the screen perceived far behind). But we will see that it is much unlike with the very large screens. By means, with a screen far away, the “scenic box“ have to be almost entirely coming out into the theatre. Anyway, If shooting have been done for a 21’ (6,5m) screen width, it will be necessary to recalibrate a "Master" – for instance for a 30/33'(9/10m) screen standard, with a 5/6 pixel shift – to comply with the far planes position in the theater. If a Master rather calibrated for a 42'(13m) width would absolutely be wanted, then a little ten pixels will have to be “dropped“ (what generally goes unnoticed). 34 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 4. SCREEN WIDTH AND MATCHING BACKGROUND OFFSETS. With large screens, background screen offsets are proportional to screen width: Screen width (meters) % Ratio 26 19.5 13 9.95 6.5 ¼% 1/3% ½% 2/3% 1% 1/400 1/300 1/200 1/150 1/100 Screen offset backgrounds measured on screen plane (centimeters) From 6.5 to 10 Fron 6.5 to 9 From 6.5 to 8 From 6.5 to 7 6.5 Pixel offset value to get comfortable backgrounds 5 to 7 pixels 2K or HD 7 to 8 pixels 2K or HD 10 to 12 pixels 2K or HD 15 to 16 pixels 2K or HD 20 pixels 2K or HD With smaller screens, background screen offsets are not anymore proportional to screen width: Screen width (meters) 5 4 3 2 1.5 1 0.65 0.50 0.30 0.26 0.13 Mobile phone % Ratio 1.10 1.30 1.60 2.00 2.27 2.50 2.27 2.10 2.00 2.00 2.00 2.8 to 3.3 1/90 1/73 1/63 1/50 1/44 1/40 1/44 1/48 1/50 1/50 1/50 1/30 Screen offset backgrounds measured on screen plane (centimeters) 5.5 to 6 5 to 5.5 4.8 to 5.3 4 to 4.5 3.4 to 4 2.5 to 3 1.5 to 2 1.0 to 1.3 6 mm 5 mm 3 mm Depending of inspecting distance Pixel offset value to get comfortable backgrounds 22 pixels 27 pixels 32 pixels 40 pixels 45 pixels 50 pixels 45 pixels 42 pixels 40 pixels 40 pixels 50 pixels 2K or 2K or 2K or 2K or 2K or 2K or 2K or 2K or 2K or 2K or 2K or HD HD HD HD HD HD HD HD HD HD HD © Alain Derobe Feb. 2009 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 35 5. Two important bounce effects are of major importance to understand what is actually different on large screens. Effect 1: As soon as the lights go out in the theatre, the viewer loses easily the exact sense of distance to the screen and recreates, when he sees a gigantic picture, a better virtual resized screen, with a more probably size, such as the 21' one (6.5m) Effect 2: With a giant screen, the choice of interaxial distance during shooting will not establish any more the window position, but will better define the more or less "expanded" or “compressed“ progression of the subject's depth scale. When the shooting have been adjusted to a 21’ (6,5 m) one, the depth scale will be seen two times deeper on a 42' (13m) screen size and four times deeper on a 85' (26m) screen. This depth expansion is perfectly admitted by the audience wishing a rather exaggerated depth/width ratio on large screens. Since the illusion of depth varies with the seat position in the theater, it means the orthostereoscopic position would be then in the first rows, close to the screen. When screen size increases, lack of respect of stereo window rule goes more and more out of sight and is more and more unnoticed by the audience. Those two effects explain why breaches to stereo window rules are well accepted for large screens and also why it makes sense to work as if the screen width was a littler one. Concerning public city theaters, where average screen sizes are between 33' (10m) and 66' (20m), it is convenient to calibrate the shooting according to stereo window requirements, corresponding for instance to a minimum 21' (6.50m) screen or a maximum 33’ (10m) one. With larger ones, stereo window error is not much noticed by the audience. On the contrary, if you would choose to work directly for a 42' (13m) screen, by using the grid at half pitch (10 pixels), or less, remember that a too distant stereo window would make shooting terribly strenuous. A so far stereo window will generate a weak depth in the incitement during shooting, to wipe out close elements. Nowadays, working systematically for a virtual 21' (6.5m) or a 32’ (10m) screen seems to be a consensual way of getting good 3D. 36 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 37 6. HOW TO BLOW OUT THE SCREEN BUT NOT THE VISION. This is an extract of Alain Derobe’s book about 3D, Copyright apply The inescapable adjustment of showing stereo 3D with screen size in theatre, sometimes caused the odd incredulous reaction is quite understandable and calls for more explanation. Many experienced stereographs – who often, have been soaked into the magic 3D mixture, many years ago – have now grappled their way to understand several “paradox” of human vision. These “specific visual rear path” now seem obvious to them but that is not the case for most cinematographers and mostly to a large audience. Common sense has a habit of believing what it sees and is frustrated when it comes to admitting that the best 3D image on a screen will require a huge jump away natural seeing. Through a poor knowledge of the complex mechanisms of vision, we sometimes don’t realise that some essential, visual functions, have to be performed, even to the detriment of real size and disposition. The main priority is that the stereoscopic double image requires no major or lengthy ocular divergence from the audience. The most popular feature films and the most restful ones for delicate eyes – and especially the least traumatic for our vision health – do not require any divergence of ocular axis. This means that our visual perception of depth – really very well behaved – is able of finding, without the slightest problem, a resemblance to stereoscopic pictures, that are far-removed from what it shows. This surprising faculty leads to the non observance of any kind of proportionality, a non-observance of proportionality in regard to reality, but also between different projections. This is fortunate because proportionality respect is impossible to attain and this is why a spectator who watches the stereo images of a given volume on a small screen, then the same images on a giant screen, feels they are seeing similar depth and similar set of items. However, it can be scientifically demonstrated that it is absolutely not the case, and that the extent of the depth represented is not only completely different, but also considerably deformed. More incredible, it has to be! If – in spite of this – the audience feels a impression of realistic depth, it is because his perceptive visual system included two, extremely powerful corrective functions: the ability to rebuilt space depending on the “most probable depth possible” and the “consistency of the size of objects and individuals”. We therefore lose any hope of finding a straightforward relation of proportionality between two different-sized projections, which can be illustrated with the following. During the editing of a 3D film, a very close-up shot of a beautiful woman smiles at us from a 30 cm wide stereo monitor. The gracious oval of her face – nearly life size- - covers the full height of the image. Then, when it will be projected in an “Imax” theater with a 30 metre screen base, we shall be delighted to see this friendly face again and not that of a giant whose 16 metre high head could lead us to believe she is nearly 100 metres tall, and whose smile would terrify both King Kong and Godzilla, clutching each other in fear. Obviously, it is our visual analysis system that performs the correction to the scale and this banality seems obvious with a flat image. 38 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo It is far less obvious with stereo 3D, because depth information may hinders the re-dimensioning of the perspective. Fortunately, the perception of depth is completely separate from the analysis of width. On a large screen, the perceptive function of the “consistent size” encourages us to mentally recalibrate the subject to its normal dimensions and at the same time, to see the whole image as if it was a lot closer to our own face, with the help of darkness. The notion of actually placing the screen further away is completely irrelevant. Beside this, on a small monitor, the correct adjustment of a 3D image with the least possible depth error, would show – for example – the eyes level on the very screen’s surface while ears and hair would be showed inside monitor’s housing. Similarly, the chin and nose would stand out towards the spectator to complete the shared volume, which is nevertheless a very flat depth in relation to reality. By the by, this kind of adjustment, which adds “presence” to such a close-up shot, is better appreciated than if the whole face would have been behind the screen plane, which is visually uncomfortable or if it would have stood completely off in front the screen, which would have caused some eyestrain. For example, we can notice an offset of 2 % between left and right pictures of a same item (6 mm measured on the screen) for a lock of hair that sticks away from the back of the head; whereas for the tip of the nose standing out forwards, the difference between the corresponding images will be crossed over about 1 % (3 mm). In fact, the rear 2/3 of the depth of the head will really seem to be beyond the screen ( regular watching distance would give about an additional 5 % of eye screen distance) whereas the other, prominent third will be showed, standing out by 2.5 % on the way from screen to eye. This is a really flat 3D depth because seen from a distance of 30 cm, it will only occupy a thickness of 2.5 cm for a face which is roughly 15 cm wide! And yet if the subject turns her head, this completely squeezed “bas relief” will look at first sight, realistic enough if not completely identical to live. Pierre Allio ironically calls this kind of depth: a “Pizza box 3D”. Obviously, the depth bears absolutely no relation to the width of the face, but the visual function that “rebuild a familiar area” flies to the rescue of a genuine cervical volume to prevent it from looking like a cushion that has been flattened by a Sumo wrestler! And so the good looking face appears to be delightfully represented rather than deformed, which is very fortunate because it would be fundamentally impossible to show the true volume of a correctly proportioned skull inside a monitor of this size. Now let’s project the same pictures in the same stereo adjustment in an “Imax” theatre, respecting their inherent off sets. In view of the new dimensions of the 30 metre screen, the lock of hair in the background would now have left and right local pictures, 60 cm apart and therefore impossible to make visual fusion for anyone in the theatre. These unacceptable images would be seen double and would stays incomprehensible. In fact, our eyes cannot diverge and it is very painful to try and visually superimpose a couple of too far-removed separated images. The eyes of a spectator sitting 30 metres from the screen would have to diverge by about 2 degrees whereas a child in the front row, 15 metres from the screen, would have probably to diverge by more than 3 degrees: And yet only a few people are able of visual divergence up to 1 degree while other ones didn’t! Defects such as this one have earned such a bad reputation for 3D in the past. Then an Imax technician would probably come to readjust the global off set of the two images and “recalibrate” them. This readjustment involve now that the whole face will stays completely in the theatre area, in front of the screen and no longer behind it. To do this, the technician has to do a The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 39 general offset of left and right pictures (in the laboratory) until the lock of hair at the back is perfectly superimposed to that it appears in screen plane, which itself is far enough away for it to be pointless to go any further. This adjustment will bring the depth position of the nose (with a difference of (2 % + 1 %) = 3 % of the width of the image, i.e. nearly one metre measured directly on the screen ( with about 60 pixels offset). This huge difference is perfectly acceptable to audience’s vision because a very large ocular convergence is easy. The nose will now appear at 90 % of the screen to spectator distance, i.e. for a spectator sitting 30 m away, something like 27 meters in front of the screen that looks like 3 meters from the eye observing it. The charming face will be represented now much deeper than wide, made oval like a rugby ball and yet we shall still perceive it as a beautifully proportioned head. This little trickery – granted we are using an extreme example – is designed to help you understand the phenomena and the rules that apply to watching stereoscopic images. The actual living depth and the picture depth represented on the screen are two, very different things and it is pointless trying to apply the same size on the screen as in the subject. The depth represented is known as the “scenic box”. It has its own rules for screening in a darkened room, especially with regard to the width of the image shown. Moreover, the scenic box has very different measurements depending on the distance and the angle between each spectator and the screen and yet each spectator mentally corrects the image to achieve a credible depth and thanks to this miracle, they all have the illusion of seeing the same kind of depth. Trying to maintain a particular measurement in the scenic box is a utopist concept. Separating the foreground area (coming out the screen) and the background area by means of accurately positioning the “stereo-window“ is only valid for a specific screen width. By means, this convergence level (screen level) becomes a non-sense when the whole volume has to appear in the theatre itself. In fact 6.5 cm on a 30 metre screen is so insignificant that sometimes we no longer take it into account. Moreover, far-off distances behind the screen no longer make any difference visually. Also, with the increase in the size of the screen, the spectator can no longer evaluate the distance and systematically feel it as it was closer. The extent and the exact position of the convergence level therefore reduce regularly in relation to the size of the screen. On the other hand, small screens are perfectly situated by spectator’s eyes, consequently they tolerate a large section of the image being behind their surface. This entails the stereo window being further forward within the volume of the scenic box, just as the convergence would have if it was closer to the camera when the shot was recorded. Therefore, simply moving a whole image sideways in relation to the other one is sufficient to re-calibrate it and this automatically positions the window (the screen level). The re-calibration of images depending on the width of the screen fortunately corresponds to the logical distance between the spectator and the screen. It is perfectly justified, for small screens, to have a greater proportion of the volume presented further back and with large screens, the scenic volume is presented much closer to the spectator so that the audience feels more involved. It is unacceptable to leave elements out of place in a direct line (in the distance), with a difference of more than the space between the eyes. If this is not readjusted, the same difference combined with the master results in varying differences on the screen, depending on the screen’s size. Re-calibrating objects in the distance by means of slightly moving the images will make very little change to the ideal position of the stereo window and will not have a major influence on any coming out of the screen. 40 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo The difference is generally small, 7 pixels (in 2K) when moving from a 10 to a 20 metre screen and this protects the audience’s eyesight with little trouble. Seven pixels opposite way, will allow one to make the adjustment for a 5 metre screen and to add the necessary depth to scenic box otherwise too far forward when no re-calibrated. We can see that with satisfactory re-calibrations, of a few pixels, we can consider that the position of coming outs the screen – and stereo-window position – are not much altered. However, diplopia and damage caused by uncontrolled divergence may have consequences on the whole show, or even cause optical medical damage. Alain Derobe, Stereograph. © Alain Derobe 2008 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 41 D. GLOSSARY The terms used for shooting are underlined. The terms concerning projection are in italics. It is important to identify them fully. Many misunderstandings and wrong meanings ensue with approximate terms. For instance, il is very important to keep the "screen offset " term for the space between corresponding left and right points on the screen and to use "axial distance" to name the space between the two optical axes. Interaxial will also be used. The average distance between human eyes (2.5" = 6.5 cm) is called interocular distance (IOD) or also interpupillary distance (IPD). When terms are common to making pictures and to their watching, like the too frequently used “parallax“ notion, we recommend you say "scenic" for the first and " pictured“ or “screened“ for the second. For instance, we can say that an actual scene scale during shooting can, depending on camera adjustments, give a compressed or expanded screen scale for projection. For projection: Scenic Box: A space represented by a volume, as it is presented to the audience's view and often much different from real space from which it comes. “Stereo real state“ means a comfort zone or the “ Maximum depth bracket” or “3D screen Space“. It is the maximum amount of deep that can be showed through the screen. It can be larger than the “scenic box”. Then, this scenic box can be a little moved more in front or back the screen plane by a general offset in post production. (depth repositioning ). The other terms can be ambiguous and may involve available screen space as well as the portion of natural space recorded. Orthoscopy: Defines the viewer's position where the screen is seen at the same field angulation than the lens used for shooting. Orthoscopy is used with flat images and does not garantee a scale corresponding to real depth in 3D. Ortho-stereoscopy: Defines the position of the one privileged viewer who sees the picture through the screen frame exactly as through a room window in natural surroundings. For this only viewer, it is the combination of an orthoscopic position and of a natural depth scale. Depth compression: When the scale is flattened. A cube seems flattened through loss of thickness. Viewers diversely perceive this distortion. Depth expansion: When the scale is expanded. A sphere is distorted like an olive or a zeppelin facing to you. Variously perceived by viewers. Offset: Space on the screen separating left and right corresponding (or homologous) points of the same subject element. It is expressed in percentage of the picture width, hence of the screen base. When the image resolution is known, it can be expressed in pixels for better post-production adjustments. Rear depth: Picture elements perceived beyond the screen. Far planes: Picture elements represented in the theater the farther away to the rear. Even with a L/R offset equal or larger than interocular distance (2.5" = 6.5 mm) or more, and even if the viewer can sense deep far planes, he will never really experience infinity. To get close to this infinity, he should be himself at a tremendous distance from the screen plane. Stereo window: 42 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo Where right and left images have no offset. It is also the borderline between rear depth and elements coming out of the screen. Screen plane: Normally, the stereo window is perceived at the screen plane (as for a flat image when left and right views are superimposed), except if , by a visible offset screen plane and stereowindow have been a little disconnected, or if a “floating window“ has been artificially planned. Floating window: By moving the frame side edges, in and giving them an offset just as any picture element, vision can be tricked and led to believe the screen plane is itself in stereo, a bit closer or farther than it physically is. This is practiced in post-production with a small amplitude, mostly to dampen the shot to shot jumps. Coming out of the screen (sometimes called theater effects or front depth): Elements that seem to appear in front of the screen. They cannot be seen outside a pyramid joining the eyes of the viewer and the four corners of the screen. The amplitude of the elements coming out of the screen, vary a lot with the nature of the image, and criteria are mostly psychological. Broken coming out of the screen effect or stereo window errors: When an element supposed to be perceived in the theater is cut off by the edge of the frame. - With medium-sized or small screens, the viewer's visual perception can then be to push back the entire image as well as a large part of the Scenic Box and lead to ocular fatigue. Abuse of window errors allowed by some is rejected by an important fraction of the public. - The larger the screen, more the viewer loses the notion of distance to the screen. He has a tendancy to return the picture dimensions back to their probable size and a virtual window is then created much closer to his eyes. In fact, with giant screens, the great majority of the picture will be displayed with broken coming out of the screen effects. With immersive setups like Omnimax TM, this can concern the entire image. Roll: Residual relative horizontal level defect of an image with respect to the other. On most correction software this can be expressed in degrees or fraction of degrees. Post-production correction might mean destroying quality. Parallax: Ambiguous! Used both in taking picture or when theatre space speaking. Unlikely it is one of the main roots of misunderstanding about 3D trade. Better to leave this term to scientific or calculation use. It can means the cause as well as the effect. Strictly, it is generate by the distance between two separate point of view such as A and B in pointing a C target. When C is the stereo window (or the screen plane) one says that positive parallax is behind the stereo window and negative parallax is in the foreground. The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo 43 Zone Industrielle 27135 Verneuil sur Avre France Tel +33 2 3232 2761 [email protected] http://www.transvideointl.com Rev 0.0 – Sept 3rd 2009 44 The Stereoscopic User Book All rights reserved © July 2009 Alain Derobe & Transvideo