User:Ishbosamiya/GSoC2019/Proposal

Cloth Simulator Improvement

Name:

Ish Hitesh Bosamiya

Contact:

Email id: [email protected]

Blender id: ​ish_bosamiya

Twitter: ​@ishbosamiya

Synopsis:

Blender has had a cloth simulator for quite a while now. It is based on a system that now needs major changes.

A lot of research has been done in adaptive cloth simulation (which is the next big step towards being able to do realistic cloth simulation in reasonable computation time). By introducing adaptive cloth simulation into Blender, we can decrease the computation time per frame, thereby leading to a better quality of simulation in the same amount of time. Furthermore, it would then be possible to build really powerful cloth production pipelines. The current cloth production pipeline is slow and requires a lot of user interaction (for example, for adding the correct level of topology and redefining the stitches each time the topology changes). With the introduction of adaptive cloth simulation, the algorithm can automatically determine the necessary topology to get the correct collisions and realistic folds and wrinkles. Additionally, this project would act as the base for future improvements, such as adding contact friction and dynamic tearing of cloth.

Benefits:

The current cloth simulation system in Blender is highly dependent on the artist creating the correct topology so that collision takes place in a meaningful manner. This should not be the artist’s job- they should not have to care about the number of polygons used, the spacing between them, etc. that currently are necessary for the accurate simulation of cloth. This project will allow artists to focus more on what’s important to them- the art. They should be able to define the overall mesh and let the computer handle the rest. Additionally, this project will change the way clothing for characters can be designed. Currently Blender ​ does ​ have stitching functionality, but it is limited- new stitching edges (loose edges are considered as stitching edges) need to be made each time the topology is changed (to ensure correct collision). This is a clunky and time consuming system that artists shouldn't need to bother with. In short, adaptive cloth simulation will allow for a simplified workflow for artists, where they just define the parts that need to be stitched, everything else “Just Works^TM”.

Deliverables:

I will implement anisotropic adaptive remeshing that would be compatible with the current cloth simulation code. I will also add some additional settings for the remeshing part of the simulation to the user interface.

If time permits (and I hope it does), I will also implement a new object type (namely, the cloth object), which will have its own interfacing options to make stitching of cloth simpler.

Project Details:

This project would add anisotropic adaptive remeshing to the cloth simulation pipeline. This will not only help improve the performance on large scale simulations, it will also make it easier for artists to use the cloth simulator. The current cloth simulator as mentioned in ​this post​ is based on a very old pipeline. The recent addition of angular bending springs did add a significant amount of realism, however, the cloth simulator is lagging in quality in comparison to competitive products.

I will base most of my work on the foundations laid by the research paper titled “Adaptive Anisotropic Remeshing for Cloth Simulation” by Rahul Narain, Armin Samii and James F. O’Brien (published at SIGGRAPH, found at this ​link​). Specifically, to implement this, it will involve adding additional steps to the cloth simulation to allow for anisotropic adaptive remeshing. The paper uses the sparse Cholesky-based solver in the TAUCS library; however, this can be swapped for the existing Blender solver code, which is what I’ll be doing.

An important point to note, in regards to this, the source code for the implementation described in the paper is available ​here​, but the license is not compatible with that of Blender’s. While this means that to add it into Blender, we will need to implement it from scratch (while referring to and citing the paper), it also means that we have a “ground-truth” implementation to compare against, in case we face any unanticipated difficulties during implementation. This makes me more confident about the timeline that I describe in the following section, as well as in the overall success of this project.

Further improvements that I foresee (for after GSoC) are to the solver (for example, to add an implicit frictional contact solver for the adaptive cloth simulation implementation as per ​this paper which would add the correct contact friction solver to further enhance the realism).

Project Schedule:

I’ve written a rough schedule below, based on my current understanding of the research paper. I will continue refining it throughout the project depending on discussions with my mentor.

Timeline	To Do
Until 27th May (i.e., before GSoC begins, and during community bonding of GSoC)	Familiarize with the current Blender cloth simulation source code Understand the research paper thoroughly
27th May (GSoC Start) - 28th June	First order approximation of implementation of anisotropic remeshing with simplified Blender cloth simulation code (remove “advanced” features and slowly add them back) Test and refine this implementation, testing done by comparing the results against the examples part of the arcsim library
29th June - 26th July	Improve implementation, while integrating it completely with existing Blender code base Ensure all existing features work as expected through integration test by plotting graph of error of distance separation between the existing high resolution cloth simulation vertices and the new integration vertices
27th July - 11th August	Optimize code to speed up simulation time
12th August - 26th August (i.e. until Official GSoC end)	Reserve time period (in case anything unforeseen happens; if not, then I’ll be adding more features or making further quality-of-life improvements) Cleaning up & finalizing project documentation

I will be updating the project documentation throughout the project, and will be finalizing it during the reserve time period.

If time permits, I will also make some preliminary steps towards the cloth production pipeline, which would involve making a new cloth object type for faster structure generation, alongside a more user friendly stitching method.

Schedule conflicts:

I have college exams during community bonding period until May 16th.

College starts again on 12th August (tentative date). I expect to be able to continue working on the project even once college starts, however I expect my timings for working to change. I have factored this into deciding the schedule above.

Bio:

I, Ish Bosamiya, am currently doing my B.Tech. in Computer Science and Engineering (2​^nd Semester) at PES University, Bangalore, India. I have delved into many different fields but design and programming have been a constant interest of mine for many years now. I first started to use Blender seriously 3~4 years ago. At that time, I made a few physics simulations, and then dived into modeling, texturing, rigging and animation. Thereafter I moved on to shading. The math behind the light calculations fascinated me, and this encouraged me to build my own ​software rasterizer​ in C++, a ​ray tracer​ in C++ and Python, and a simple ​2D physics engine​ in C++ to gain a better understanding of the computer graphics world and further my programming skills. Recently, I have been spending my free time to add new and cool features to Blender. A link to my profile can be found ​here​. My latest addition has been to the Triangulate Modifier, to allow users to have better control by adding a minimum number of vertices parameter to it.

Working on a physics engine from scratch has taught me a lot about the complexities of simulations. The different patches that I have made for Blender (some already merged into master,and a few more in progress!) have let me get accustomed with Blender’s code base, which means that I will hit the ground running. Since I already have experience in implementing simulations etc. from scratch, I have a good idea of the specific complexities and issues that may come along the way. In addition, I work well, both independently, as well as in a team (or with a mentor). I am strongly motivated, and want to build the necessary foundation to build a great cloth production tool in Blender in this year’s GSoC; and make regular improvements to have a tool that is extremely powerful and robust. All of these make me the perfect candidate for this project.