Conclusion


   The aim of this project was to create an animation that looked like a moving Vangogh painting. To do this I had to devise a way of rendering 3D images in a non-photorealistic fashion. My technique had to be robust and error free so that it would generate consistent images of consistently high quality.

   The technique I chose relied on five pieces of software. Therefore my entire project was dependent on the reliability of these programs. Thankfully I did not have any major problems in this respect. For instance with Maya and Renderman I did not encounter any problems that could not be solved with a bit of thought. I think this is because I now understand how these programs work - much more so than I did last year.

   If I look at each aspect of this project I am very pleased with the results. The images generated actually do look quite painterly and certainly resemble some of Vincent Vangogh's paintings. I am pleased with the way I broke down Vangogh's work into a series of characteristics and then attempted to replicate these in the computer. I think this methodology worked very well and made my task more manageable.

   The modelling was not particularly demanding and was completed quite quickly. I admit that I could have fleshed out the environment with more objects and buildings but this would not really have improved the likeness of my work to a Vangogh painting - which was my aim. Indeed many of Vangogh's paintings were quite sparse. Most of the objects were of the correct geometric complexity aswell - i.e. not too complex and not too simple. The lighting was a bit more difficult to implement but was just as successful. In truth the lights played a smaller part in generating the final look of the animation than in a photorealistic project. A lot of the illumination of objects was generated within the Renderman shader.

   Which brings me on to the most important part of the project. I had only written one or two very simple shaders in the past and so felt a bit daunted by the prospect of writing a painterly shader. I bought myself a copy of the 'Advanced Renderman' book and read the whole thing from front to back. By doing this I discovered that there was a way of accomplishing the task with textures and a few clever tricks, rather than in a purely procedural and mathematical way. Hence my final solution. Programming has never been a problem for me but maths has, so I focussed on what I was good at. This seemed to pay off. Infact I was very suprised at how smoothly writing the shader went. I quickly came up with an idea that worked and looked good.

  Probably the best way of analysing how successful this project has been is to compare it to some existing work:

 

My Old Vangogh Project

   Back in 1998 I produced another Vangogh animation as part of a group project. I have been comparing the two recently and have been astonished at the differerence. My new work is so much richer and detailed than the old project. I attribute this improvement to the greater variety of software and tools at my disposal and to the development of my own skills. For instance the old project did not use any type of compositing, colour correction or custom shader.

 

'The Painterly Approach' (As Outlined in the Paper)

   The images generated by this technique appear to be very painterly. They differ from my project because the technique uses textured particles which are applied to the surfaces of the objects. This gives a rougher, more random appearance to the images. I think it fair to say that for a single image this papers technique seems to produce slightly better results than my method. However it has two major limiting factors. Firstly it only works within the custom software developed for the project. My shader works with Renderman which is a widely used program. Secondly I would hasten to guess that the technique in the paper is quite slow to calculate due to the huge number of particles that must be needed for a complex scene. My technique renders images very quickly - although a compositing stage is required.

   It would be interesting to see the 'painterly approach' applied to an animated scene. Applying particles to objects may cause problems if the objects are in motion or deforming. My technique works in both scenarios.

 

   Regardless of how good my method was, it is the final rendered footage which is judged. I am pleased with how this has turned out. I was originally going to include six shots in this project, but in the end two was a more feasible number. I spent a considerable amount of time polishing these shots and removing all the errors. Some of my previous work has had small glitches which can be noticed during the animation, but this project has none what-so-ever. Even studying the footage on a frame by frame basis reveals no mistakes. This robustness is another way in which my work has improved.

   The additional components of this project such as the flocking crows and compositing were completed as mini projects on the side. To do this I had to learn how to use Chalice and Houdini. I found Chalice very easy to pickup and enjoyed the compositing part of this project. I have gained a firm understanding of this 'image processing' side of effects. Houdini was more complex and had a far steeper leaning curve. I just learnt enough about the particle editor and object editor to complete this project. 

   No project is perfect and there are numeous improvements I'd like to make to this one. I would like to complete more shots and add more buildings. I would like to add some animated people and perhaps other animals. I would also like to improve the flock of crows (as outlined in the flocking chapter in this report). The greatest improvement I could make to this project would probably be to increase the texture map resolution. The ones I used were 400x400 pixels but I would have liked to use textures three or four times this size. It is a well known fact that for films and even television very high resoultion texture maps are used to maintain detail at close quarters (but this would have slowed down rendering too much at university).

   Overall, given that I had only three to four months to complete a project like this on my own, I think I have been very successful. Hopefully, with experience, my work will continue to improve.