troygbiv

This article is sort of a primer on my experience thus far with Vray and its varied intricacies. Based on my knowledge currently I would also like to share some opinions on differences I noticed between Mental Ray and Vray. There are quite a few things I actually find rather intriguing about Vray, being a Mental Ray user for some time now, I am surprised some of the functionality in Vray hasn't been adopted in some form or another inside of Mental Ray. I suspect however based off of some of the newer technology changes we should see some major overhauls in Mental Ray as a whole towards Unified sampling and full BSDF support which I intend on studying more in the near future.

Inevitably this sort of turned into a long technical discovery on the methodology and implementation experienced when rendering with Vray. Of course covering every aspect would probably produce quite the document to read so I decided to hit up areas of particular note. Hopefully I will have more notation as I move along and add more individual projects or work I have had the priviledge to be a part of.

I suppose we can start this story like any other, surrounding the depth of the sampling algorithms implemented in Vray versus what is commonplace inside of Mental Ray and other similar rendering engines. Even as I type I am still trying to fully wrap my head around the sampling settings inside of Vray, there is a lot of flexibility and having the opportunity to use it will definitely help to hone the necessary samples required on a case by case basis in future endeavors. Mental Ray for example if you are unfamiliar with it uses an adaptive subdivision method taking samples at the corners of each division based on the contrast threshold of the color in the scene up to a limited threshold amount designated by the user. However, you can jitter sampling outside the normal edge of each pixel using Mental Ray which provides a random sampling, ultimately its algorithm is still at heart a subdivision sampler. This algorithm is as well an included sampling type in Vray. The method to which Vray subdivides the scene is handled by creating a grid that is placed on top of the pixels in the camera and subdivides the sampling based on the threshold discussed earlier, disadvantageous because of the memory requirement to store data from the grid as it samples and converts the output to a raster image.

But fear not as it turns out Vray also has its own methodology for executing sampling by adaptively raytracing samples from the camera not specifically hindered by the grid like limitations of subdivision sampling. This method is referred to as Deterministic Monte Carlo or more affectionately DMC for short. For an educational primer on Monte Carlo I would refer you to the Wiki page MonteCarlo. Based on my understanding, while using Adaptive DMC the initial sampling control derives itself from the Adaptive amount settings, which actually exist separate of the Image Sampling options. But before we go into that lets get a correct frame of mind. Tackling this intellectually has its benefits but framing ones self prior to exploration helps to determine the comprehension of the subject. Lets assume above all else that there is no 100 percent unification of sampling. Granted all of these settings work together to provide some unified output but breaking this down into components has its advantages. Hypothetically we assume that setting sampling amounts on these individual elements in this case Materials or Lights dictate the total amount of allowable sampling per element. We setup a scene and provide each element with a number we would consider to be appropriate. However we are never truly going to be able to recognize the maximum sample scope which is why we still require the need of a threshold of sorts to help guide the sampling engine in the direction that is accurate.

After designating the appropriate numbers of sampling for each element we would then determine the minimum and maximum sampling amounts as a whole. This initial measurable amount is taken from the "Adaptive amount" parameter, 1 being fully adaptive and 0 being non adaptive. Hypothetically speaking if we set the subd on a material to be 50 the Adaptive amount would specify when to start adaptation. By setting the option to .5 we are sampling 50 percent of the specific subd set in each element and then transitioning to fully adaptive after those samples have been cast. The minimum samples parameter acts as sort of a fail safe for the minimum required samples Vray must cast before moving on to adaptive methods. For instance if we had set the Adaptive amount to 1.0 and the minimum samples attribute to 1 we take a minimum sample of 1 and then allow Vray to determine based on that sample whether it requires more samples to clean up noise in the image. This can be obviously non beneficial as there may not be enough information to make an intelligent decision in just one single sample. Based on this you can already tell there is a larger range of complexity in determing the outcome of the image in comparison to the more widely used subdivision method. If you would like something to really blow your mind imagine for a second that the adaptive subdivision is directly correlated to the material subdivision sampling, if you were to cast 200 adaptive eye rays into a scene and the material had 200 subdivision samples you would only get in return 1 sample for each eye ray for reflections. But setting your render at 10 eye rays and 200 material samples is not always intuitive either, sometimes geometry will be much more complex than a simple floor. In this case you will need more initial samples to resolve the geometry itself. So how the heck do we ever know exactly what is necessary to sample and what is not? Good question. Ultimately its up to the individual and their needs when render time comes around.

But it should be stressed that there is a method to unify most of the sampling done in camera to achieve a final result. It is based on the writing of Robert Nederhorst  NederHorst Primer. Basically he changes the settings to fully adaptive at which point you would merely control the sampling through the Image Sampler and not in the DMC sampler. He also has other information written up about Global Illumination preferences, however, most of the other information is relative to the product you are outputting.

One of the primary aspects I really enjoy so far with Vray would be the Physical Camera system which mimics the way a camera would work relative to rendering based on the real world parameters that are supplied. In turn you can achieve aesthetic results through the lens as long as you are aware of placement and settings and then adjust accordingly. Sadly there seems to be no robust tone mapping control. With Mental Ray you have extensive control within its photographic shader, which allows you to tone map and adjust the image inside of the render. This is a definite benefit to anyone looking to output a single render with little to no post composite work. Which in my personal opinion is a nice advantage.

In the next few articles I plan on potentially setting up a render by integrating a vehicle into a live action sort of plate, as well as potentially other various objects provided I have the time to expand on the content.