Initial Particle Velocity in the Unknown Initial Position ablation

[Rabona17]

Hi Shanyan,

Thanks for the nice work!

I noticed that the transition model needs the initial state, and in Table 5 you showed results for estimated initial particle position. I am wondering that, in the case of estimating the initial position, how to you get the initial particle velocity to feed to the model?

Thanks

[syguan96]

This is a good question. This paper didn't give a solution to estimate initial velocity. I think a feasible solution is to estimate it from a short clip at the start of the video.

[Rabona17]

Thanks for the quick reply!!!

I see, so how was the ablation on UNKNOWN INITIAL PARTICLE POSITIONS done? Did you estimate the initial velocity?

Thanks again

[syguan96]

Sorry for the late reply! I'm a little busy these days.
For UNKNOWN INITIAL PARTICLE POSITIONS, we use Nerf and marching cube to extract the initial mesh from multi-view images. However, we still use the gt velocity. In our dataset, the velocity at t=0 is 0.

[syguan96]

The requirement of initial particle velocities is one of the limitations of our paper, which has been discussed in Section 6.