Advanced Production (Summary)

Despite returning to Semester 2 later than expected (due to illness), I have made excellent progress, with major breakthroughs in overcoming many of the challenges my project was facing.

My programme of study takes a more visualisation based route, and I am working with the University mathematics division to visualise numerical data (gained from their mathematical models) which represents the growth and development of cancer cells (or solid tumour growth).

Working with numerical data has created it's own set of challenges, as Maya cannot read or interpret this in it's off-the-shelf form. As a result of this problem, I have developed skills in using Maya's own scripting language, MEL (Maya Embedded Language) and also a second scripting language (used more widely) called Python. Although this has been a difficult and time-consuming process (several months of learning, which is still ongoing) I have been able to harness this new understanding and create custom tools which can be used to read the mathematical data and create appropriate 3D geometry.

Throughout learning each scripting language, I referred primarily to online sources, and the 'official' documentation provided. I also used the Digital Tutors service, which was both informative and relevant, helping me overcome the hurdle of knowing where to start, and what to focus on.

Through creating my own custom tools, I was able to work with the numerical data-sets provided. Starting with simple tests, I was (eventually) able to construct fully animated scenes containing simple spheres, representing the individual cancer cells, and although time-consuming to process, these results opened the way for more advanced development of the data and it's visual attributes.

I spent some time experimenting with Maya's render layers, and compositing passes using Nuke. This was as a response to feedback given last semester - not everything has to be done in 3D, as it can often be quicker and easier to complete some work in 3D, but fine-tune the details in post-production. This helped me develop possible ideas for the look of cells, when considering texture on a microscopic basis, and I also experimented with 'faking' depth-of-field techniques to enhance this.

Although creating spheres was great, I wanted to push my skills further. I spent time working with RealFlow, Cinema 4D and Houdini - three great 3D packages. With the exception of RealFlow, these were entirely new to me, and although Cinema 4D was fairly straight-forward, Houdini had a very steep learning curve. Using my Python scripting skills, I created a large-scale metaball system, equivalent to the spheres created in Maya - this was a challenging task, forcing me to solve several problems, whilst trying to script in a language mostly new to myself. The result of this technical experimentation allowed me to create an interesting almost-organic 3D structure, which moved and behaved as one surface, instead of 1068 individual cells.

After completing these Python related tasks, I returned to Maya to work with additional data-sets, still showing the growth and development of cancer cells, but in a different format. This data would require not only animated movement of cells, but changing colours, and eventually, complex density 'clouds'.

The difficulty at this stage in my work, was the size of the data-sets... one of which contains over 30 million lines of information (a text file weighing in at around 800mb). Despite breaking this into smaller chunks of data, it was still computationally intense, and difficult to work with (often crashing high speed computers). I spent time streamlining and optimising my scripts, and the way they handled the data - as an example, my oxygen density script was originally taking around 350 seconds per frame to process, whereas now it takes around 90 seconds, about a quarter of the time. As I have learned more about MEL, and gained a better understanding of the language, I have been constantly improving and refining my scripts, to ensure the best efficiency when working with large data sets.

I am currently working with new additional data-sets, which again have constantly required me to think and problem solve... now that my scripts are optimised, I am confident that my scripting abilities are suited to a wide range of data types, across two very different scripting languages. This allows me to tackle upcoming problems with new knowledge, based on the experiences I have had over the last ~3 months, giving me a definite headstart.

In summary, my practice-based learning has developed the following;

• Scripting - both MEL and Python, for simple and complex tasks (including 3D visualisation)
• Improved software knowledge - Maya, RealFlow, Nuke
• New software knowledge - Cinema 4D, Houdini
• Advanced 3D skills development - particles/dynamics, data input/output, script optimisation, rendering
• Experience in 2D compositing with 3D-based image sequences

Due to the sensitive nature of the mathematical data I am working with (and as the source data is effectively the result of unpublished research) I am unable to post a large amount of my visual work online. However, this sensitive material will be made privately available during course presentations, and to module assessors as necessary.

Snake In The Grass

...more specifically, a Python.

Continuing on from my Cinema 4D experimentation with metaballs, I abandoned that line of testing, as it wasn't going to be a viable option for the large amount of cells I was dealing with. Alongside Digital Tutors, I returned to Python scripting, trying to 'translate' my MEL script into something useable inside of Houdini.

After almost a week's worth of scripting, fixing things, and re-scripting, the Python/Houdini version of my data-loading script works, creating all of the necessary nodes in Houdini, and keyframing all of the animation. This allowed me to natively create metaballs, rather than try and convert existing scene information. Houdini handles metaballs exceptionally well, adjusting the viewport geometry on the fly (so it doesnt crash regularly like Cinema 4D). Using Mantra, I was able to render out the full 1900 frame image sequence, with frame 1900 shown below;

Although I am still learning to use Houdini, I have made breakthrough progress in using the software alongside Python, opening up a whole range of new opportunities. I have started trying to figure out shaders and lighting in Houdini (which seems more difficult than expected), and I created some example 'looks' for my meta-surface, shown below;

I am due to meet with the mathematics department this week, where I will present all of my research, ideas, and generated media - bearing in mind, that they have not seen any of the results so far. I am hoping that this meeting will help inform my next steps, and direct my technical understanding to a new visual solution.

Reaching Limits...

This week has proven difficult in terms of my ideas being restricted in their implementation, primarily by the software/technology I have available.

Scripting has made a reasonable amount of progress. I have added colour changes to my first data-set, as it was not clear when cells 'appeared'. New cells begin red, and fade to their regular colour over 25 frames (1 second), which is more informative, and better communicates the data visually. It is important to point out here, that this change was added to the MEL version of my script. A render taken from the Maya scene can be seen below;

I started developing skills in scripting using Python, as Houdini relies on this for data input. Python also crosses over with Maya as the shared 'language'. I began by trying to find the common links between MEL and Python, which accelerated my understanding of how to code effectively in Python and make things happen. The Maya-Python version of the script is now equivalent to the MEL script.

However, the Houdini-Python script currently only reads data, but does not yet generate or animate geometry - I am having difficulties finding useful information on the Houdini specific Python module (commands), so this part of scripting is on hold for now.

Due to these difficulties, I returned to Maya, and started work on trying to create an organic-looking cell surface - that is, a single surface which is created from all of the cells, and moves and acts as one (instead of 1067 individual cell 'spheres'). Initially, I wanted to generate a particle field in place of spheres, but, unfortunately, Maya does not allow transformation of individual particles within a particle object. There is no way around this, except by creating a separate particle object per cell - by doing this however, the particles can no longer merge together as one.

I started to consider other options to achieve this organic look, with RealFlow being top of my list. I generated a 'particle sphere' in Maya and exported this directy into RealFlow. After developing a complicated workflow of imports and exports, and a lot of experimentation, I was able to then use RealFlow to mesh this particle field. However, for one cell to be meshed fully, I required around 37,000 particles, generating approximately 110,000 faces for the mesh. With a scene containing 1067 cells, I realised that although giving a nice result, it was certainly not a viable option. A render of the style of cell-split can be seen below;

Beyond using RealFlow, my next efforts will involve using particles or metaballs in Houdini. Metaballs work very nicely (tested using 2-3 objects manually), and are 'clever' spheres which merge together when they are close to each other (without any configuration in Houdini, metaballs give a great looking result). Particles in Houdini will hopefully offer more flexibility than Maya, allowing me another technique to create the look that I want to achieve.

Although I have found this week frustrating, finding limitations in the software I am using, I am confident that I will be able to find a solution, allowing me to realise the full potential of the creative ideas I want to unleash...