Possibility of supporting per-frame obstacle

[williamljb]

Possibility of supporting per-frame obstacle

[williamljb]

Hi xiangdonglai,

The current version does not support this. However, it only requires very few changes. I posted a version of the simulation code here, where it is used to generate synthetic data in my other project. The major difference will be in arcsim/src/obstacle.cpp, where I changed the obtacle I/O to support per-frame obstacle. There are also other minor engineering changes such as flags/config params, etc.

I think these changes can be directly applied to the differentiable code, as long as the obstacles are not used for gradient computation.

[xiangdonglai]

Thank you for your quick response. I am wondering what you mean exactly by "the obstacles are not used for gradient computation". For example, is it okay if we do not require the gradients with respect to the obstacles, but the obstacles still influence the cloth during simulation?

[williamljb]

Oh, what I said was actually not correct. On the one hand, since the obstacles affect the cloth motion, their values are indeed in the computation graph (what I wanted to say is that they are basically constants). On the other hand, if you want to optimize the position of the obstacles, it is also possible by setting it as a variable, although you may have to save the mesh to the hard drive for each iteration, which is only a matter of implementation. Sorry for the confusion!

[xiangdonglai]

I see, thank you.