The original article this has been modified from can be found here.
In part 1 of this study, it was discussed how this project was created to test the Mental Ray rendering engine’s ability to create a viable lighting simulation. As highlighted, a photograph of an existing reconstructed structure at the site was initially provided as a reference to match (below). In previous posts it has been discussed how the basic lighting was setup, specifically in relation to the use of a daylight system, final gather, photons, ambient occlusion and participating media.
Photo reference from Çatalhöyük
Creating the model
The photograph above clearly shows the areas of the image that needed to represented. Contrasted, dark shadows cling to the walls and baskets at the right wall, forming a very distinctive falloff of light and likewise spilling into the room from the roof entrance. As this project was primarily aimed at producing accurate lighting, much of the focus went into creating the geometry and matching the perspective of the imagery to really make sure that the light had a similar environment to enter and move around. With anything that is related to lighting with 3D graphics, providing the correct scaled dimensions is critical and often models will provide strange results due to the hidden inconsistencies that arise from confusing this setting. This should not only the first priority that should be considered, but also one of the most critical in creating any model that aims at achieving a convincing light falloff.
Without any blueprints/dimensions and only relying upon a solo photograph, it was a difficult task to create a model with ‘true’ accuracy (Or at close as you can get with approximated techniques), but restrictions of this nature are common in the creation of 3D content and should not be seen as a reason to avoid modelling. Often software provides photo-matching help to establish perspective, especially from photographic content, but for this specific project I manually matched the shot, to gain a better understanding of the composition and the angle that I needed to reference. Furthermore, as with most automated systems, they are never reliable in every situation and problems frequently arise, so gaining practice at manually overcoming this problem was welcomed. Fortunately within this shot some information was provided in the metadata of the photograph (ISO, shutter speed, aperture & lens settings), which allowed a mirrored camera to be created within 3DS Max. Draft geometry was then created to represent the walls and roof, which was tweaked until it lined up closely with the perspective of original image. The lens information made it a fairly straight forward process to achieve a decent representation of the perspective and likewise, knowing the exposure settings provided a great starting point for the composition.
In relation to the materials, everything used in the scene was an arch and design shader, as these are specifically designed to be energy efficient and work within the confines of the Mental Ray rendering techniques. Texture maps were created in Photoshop from a photographic archive from the site and made tile-able using the clone/offset tool to remove any seams. Examples of some of the references can be seen below.
The majority of the objects in the scene were created using very basic geometry creation (As can be seen below), with the exception of the baskets that were made using renderable splines, but as the main focus of the model was to study the lighting, there was not really a great deal of time investment in the shaders or objects outside of creating something that was recognisable in size, shape and placement within the model.
Wireframe of the shot
An example of a renderable spline basket
Analysing the lighting
This is the image that combines both final gather and global illumination. It provides a good accentuation of the textures and depth to the shadows. If there is one criticism, it is that there is an over exaggeration of colour bleeding into the room. It feels too warm in relation to the photograph, but this is most likely just an issue with the GI settings.
This image is much cooler in its complexion and therefore closer to the original in its colour. However, only using Final Gather has made the interior surfaces seem too smooth. In comparison to the image above, there is some contrast missing to the image, especially in regards to lighting around the lip of the roof entrance.
Light Spill 01 – Left Photo Mid FG/GI Right FG
Light Spill 02 Left Photo Mid FG/GI Right FG
Light Spill 03 Left Photo Mid FG/GI Right FG
It is clear from looking at the images that Mental Ray has produced a good series of images that reflect the photograph very closely. When combined, even with very rudimentary geometry, textures and without specific measurements, the series of images are comparable to the original. The ambient occlusion has also brought out the shading around the baskets and in the corners of the room successfully and if anything is a little bit too intense. The light penetrates the area between the baskets in exactly the same manner as the photograph, leaving accurate arcs of shadow where they block the wall and the spilling of light is almost identical in the renders, with the volumetric lighting provided the added extra realism needed in representing the gentle haze and bloom.
Final colour corrected comparison
To finally bring the result to a close resemblance of the original, a little bit of post production was used to colour correct and manually tweak the levels and curves of the image. Using post production in this way is a very common process, especially for architectural work and whilst this image had minimal editing, often the amount of editing that occurs in visual images is heavily underestimated. If anything the colour in this image above is a little to green and the shadows are general darker than the photograph, this is a result of the material properties not providing enough bounced light through reflection.
Looking at the final output images, it is clear that there is great potential in the Mental Ray rendering engine to deliver clean, convincing imagery and as an out of the box renderer within 3DS Max, it is very good. However, I hesitate to use ‘reliable’ as a definition simply because although physically the way the light is falling is realistic, it is also an approximation. Furthermore, although each render has slight differences (The GI/FG vs FG only), they both fulfill the claim by van der Steen (2007) that they are ‘physically accurate’, yet they present different visual results. In the context of computer graphics something being a ‘physical accurate’ method, does not translate to, ‘every image that is produced in this way is definitive’, yet this is often how this statement is taken and spun for the purpose of authority. Also, as seen above the manual tweaking within Photoshop to enhance the image with what are in effect subjective, artistic techniques can still be put forward as being ‘physically accurate’.
Ultimately, achieving a good representation of the photograph has not been an automatic process. The model forced the trial and error testing of different lighting setups, materials and the recreation of geometry that did not fit or project the right ‘look’. This was a very manual, subjective workflow and achieving the final result was more down to a gradual improvement of user knowledge rather than the mathematics of a machine. Being able to justify using this process at Catalhöyük, where a synthesis of virtual reality could present art and culture in a idealised habitat, is obviously very important as it could provide a new dimension to both research and viewer experience. This is not just critical from an aesthetic and luminary view, but also when taking into account how light affects both perception of space and such cultural inferences as art, religion and the domestic household at the site. However, this process has shown how much of using 3D packages like 3DS Max/Maya is directly down to the users control, experience and knowledge of the process. Consequently, it would be possible to go back to this model and produce a whole series of images with tweaked settings, materials and geometry and produce something that whilst looking similar to the images above, is different and even if you were to take away the lighting techniques (FG/GI) and replace them with a more automatic system (such as an unbiased renderer for example) there is an incredible amount of depth and knowledge needed to create imagery that is interchangeable with reality. There are just far too many variables at this moment in time to treat computer graphics as an automatic process and this is without even considering constraints like hardware, software, budget, time and the chosen workflow.
The techniques shown, such as ambient occlusion, participating media are a snapshot of the ever growing arsenal that is available to a 3D artist and these techniques are not only the tip of the iceberg, but also a homage to the development of graphics throughout the last few decades and they represent a very large network of processes that fit into studio and individual workflows around the world. You only need to look at the hours of specialist tutorials from an advanced user such as Grant Warwick to understand exactly how valuable and rare it is to find artists who can truly produce images from scratch that mimic reality to the point of photorealism. This is because each texture/shader/material is telling a story in the way it is created and to do this users need to build up the necessary mind frame to think in this way and deconstruct physical phenomena and objects. This actually makes creating virtual models very important for the conceptualisation and content of archaeological material because it forces the user to be able to analyse space, light and material properties in a much more diverse way. For this reason alone it is felt that this study has been a success, as it has opened up a new practice of thinking and applying this to structures that no longer exist is an exciting prospect.
So, consequently whilst this model definitely does not achieve photorealism with its simple geometry and materials, as a sandbox for thinking about archaeological content, Mental Ray has proved an immense amount in this study. Not only has shadow and light has been convincingly represented, which fulfills the aims of the project, but also the groundwork for conceptualising physical culture at the site using tools that allow for the deconstruction of reality has been laid.
Van der Steen 2007 Rendering with Mental Ray