Fusion4D: Real-time Performance Capture of Challenging Scenes

Рет қаралды 70,100

8 жыл бұрын

We contribute a new pipeline for live multi-view performance capture, generating temporally coherent high-quality reconstructions in real-time. Our algorithm supports both incremental reconstruction, improving the surface estimation over time, as well as parameterizing the nonrigid scene motion. Our approach is highly robust to both large frame-to-frame motion and topology changes, allowing us to reconstruct extremely challenging scenes. We demonstrate advantages over related real-time techniques that either deform an online generated template or continually fuse depth data nonrigidly into a single reference model. Finally, we show geometric reconstruction results on par with offline methods which require orders of magnitude more processing time and many more RGBD cameras.

Пікірлер: 26

@dr-maybe 8 жыл бұрын

You guys are doing astonishing work!

@RuofeiDu 8 жыл бұрын

This is awesome work! Congratulations to the MSR I3D / Holoportation team!

@KraitoKrombongus 8 жыл бұрын

the potential of what you can create from this is sky high.

@ScientificGentlemen 8 жыл бұрын

The future of Cinema, theatre, T.v., Video Games...you name it

@3Daver 8 жыл бұрын

Absolutely brilliant tech! Not so much for film or games because the money is not in the capture, but in the retargeting. This, however, could get you a VR seat at a show on Broadway or The Superbowl. Amazing.

@Spieluhr009 4 жыл бұрын

Excellent performance!

@sl4gg 8 жыл бұрын

this is gonna revolutionize motion-capture

@dbcomix 8 жыл бұрын

Especially with independent film makers/ game designers.

@do__ob 8 жыл бұрын

This must require one hell of a setup to work in real time with minimum lag.

@oBCHANo 8 жыл бұрын

This captures depth and uses it to morph a volume that is then converted into polygons. As far as it's concerned each frame is just a static mesh. There is no information that can be applied to a rig. This is good for capturing big events for VR, so that you can allow people to move around and maintain true perspective, imagine hundreds of these set up at a football stadium and then have it rendered with lightfields for VR, that is what this kind of tech is for, it doesn't really benefit motion capture at all.

@os3ujziC 7 жыл бұрын

Amazing stuff.

@SuperTubeAndi 8 жыл бұрын

Awesome - hope this will be made available publicly to use/license/implement. Do you include approaches to deal with reflective/refractive/transparent surfaces or absorption? How about reconstructing from RGB(without D) cameras?

@TheGamingMackV 8 жыл бұрын

If this revolutionizes motion capture, does this mean less work to put into making a 3D model? Such as designing the head, facial movements, mounting the 3D modeled head onto a motion captured body, ect?

@VijayNirmal 8 жыл бұрын

great

@creeperlamoureux 7 жыл бұрын

4:25 oh uh goodbye then

@zhangxaochen 8 жыл бұрын

How is the "Qualitative Comparison" carried out? the source code of "DynamicFusion" seems not publicly available till now, did you priviately acquired the sourcecode or implemented it on your own? Thx for reply~ :-)