It seems like this paper[1] lays the groundwork and the discussion in the OP is about refinements to the method.

[1] https://arxiv.org/pdf/2406.09733

I'm confused why you would think so. Did you mix up your links? The paper you link to replaces the quick-and-dirty 3D-to-2D approximation used in Gaussian splatting with more physically accurate rendering using Monte Carlo path tracing. FreeTimeGS uses Gaussian splatting for 3D scenes with movement.

Both descend from (and cite) 3D Gaussian Splatting for Real-Time Radiance Field Rendering https://arxiv.org/abs/2308.04079 but take it into entirely different directions.

"the best way to get the right answer on the internet is not to ask a question; it's to post the wrong answer." -- Ward

With the "mild" side effect of spreading lies and misinformation.

You can play with the output https://www.4dv.ai/en

Apparently using PlayCanvas.

i hate to be that buy... but the adult video industry is sleeping if they dont implement this lol

Now add VR into the mix...

People often say this about vr, but I think the truth is that consumers of adult videos are not motivated enough and the production costs outweigh the benefits. The demo scenes here were each captured on about 20 cameras, each carefully synchronized and rigged to be out of each other's lines of sight. Add the expertise and time to train the models (still more like pets than cattle) and we're getting into movie ticket territory and away from tube site

So, what you're saying that there's a a business opportunity not only on the software service side, but the logistic/equipment side as well!

> and rigged to be out of each other's lines of sight

I think there's a misunderstanding of the industry here, if you think the viewing audience will be concerned about some poorly disguised cameras at the edges of the scene.

The ones paying $20 for the viewing? I think so.

How does one use something like this on the dataset side?

tl;dr anyone ? Does this repo allow also own processing ?