Instant-NVR: Instant Neural Volumetric Rendering for Human-object Interactions
from Monocular RGBD Stream

4D modeling of human-object interactions from monocular RGBD camera.

CVPR 2023

Yuheng Jiang*, Kaixin Yao*, Zhuo Su, Zhehao Shen, Haimin Luo, Lan Xu

Paper Sequences(coming soon)

Convenient 4D modeling of human-object interactions is essential for numerous applications. However, monocular tracking and rendering of complex interaction scenarios remain challenging. In this paper, we propose Instant-NVR, a neural approach for instant volumetric human-object tracking and rendering using a single RGBD camera. It bridges traditional non-rigid tracking with recent instant radiance field techniques via a multi-thread tracking-rendering mechanism. In the tracking front-end, we adopt a robust human-object capture scheme to provide sufficient motion priors. We further introduce a separated instant neural representation with a novel hybrid deformation module for the interacting scene. We also provide an on-the-fly reconstruction scheme of the dynamic/static radiance fields via efficient motion-prior searching. Moreover, we introduce an online key frame selection scheme and a rendering-aware refinement strategy to significantly improve the appearance details for online novel-view synthesis. Extensive experiments demonstrate the effectiveness and efficiency of our approach for the instant generation of human-object radiance fields on the fly, notably achieving real-time photo-realistic novel view synthesis under complex human-object interactions.


We adopt a tracking-rendering mechanism for our Instant-NVR. The tracking front-end provides online motion estimations of both the performer and object, while the rendering back-end reconstructs the radiance fields of the interaction scene to provide instant novel view synthesis with photo-realism.

Rendering Pipeline

For the rendering back-end, we adopt a separate instant neural representation. Specifically, both the dynamic performer and static object are represented as implicit radiance fields with multi-scale feature hashing in the canonical space and share volumetric rendering for novel view synthesis. For the dynamic human, we further introduce a hybrid deformation module to efficiently utilize the non-rigid motion priors. Then, we modify the training process of radiance fields into a key-frame based setting, so as to enable graduate and on-the-fly optimization of the radiance fields within the rendering thread.


Here are our rendering results. Instant-NVR achieves real-time photo-realistic novel view synthesis under complex human-object interactions.