# MoE3D: A Mixture-of-Experts Module for 3D Reconstruction

Source: [MoE3D: A Mixture-of-Experts Module for 3D Reconstruction](<https://www.sota2.com/research/paper/69608b7953fdd5c36174446e>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-01-28T05:00:19.604Z

Authors: Zichen Wang, Ang Cao, Liam J. Wang, Jeong Joon Park

Published: 2026-01-08T18:33:52.000Z

## Abstract

We propose a simple yet effective approach to enhance the performance of feed-forward 3D reconstruction models. Existing methods often struggle near depth discontinuities, where standard regression losses encourage spatial averaging and thus blur sharp boundaries. To address this issue, we introduce a mixture-of-experts formulation that handles uncertainty at depth boundaries by combining multiple smooth depth predictions. A softmax weighting head dynamically selects among these hypotheses on a per-pixel basis. By integrating our mixture model into a pre-trained state-of-the-art 3D model, we achieve a substantial reduction of boundary artifacts and gains in overall reconstruction accuracy. Notably, our approach is highly compute efficient, delivering generalizable improvements even when fine-tuned on a small subset of training data while incurring only negligible additional inference computation, suggesting a promising direction for lightweight and accurate 3D reconstruction.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Geometric boundary sharpness on NYU v2](<https://www.sota2.com/research/sota/geometric-boundary-sharpness-on-nyu-v2.md>): Geometric boundary sharpness; NYU V2; mIoU: 19.4.

- [Geometric Boundary Sharpness on Sintel](<https://www.sota2.com/research/sota/geometric-boundary-sharpness-on-sintel.md>): Geometric boundary sharpness; Sintel; mIoU: 19.4.

- [Geometric boundary sharpness on NRGBD](<https://www.sota2.com/research/sota/geometric-boundary-sharpness-on-nrgbd.md>): Geometric boundary sharpness; NRGBD; mIoU: 40.2.

- [Video Depth Estimation on Bonn (AbsRel, δ \< 1.25)](<https://www.sota2.com/research/sota/video-depth-estimation-on-bonn-absrel-1-25.md>): Video Depth Estimation; BONN; AbsRel: 5.3.

- [Video Depth Estimation on Sintel](<https://www.sota2.com/research/sota/video-depth-estimation-on-sintel.md>): Video Depth Estimation; Sintel; Delta Threshold Accuracy (1.25): 67.7.

- [Video Depth Estimation on KITTI (Abs Rel, delta \< 1.25)](<https://www.sota2.com/research/sota/video-depth-estimation-on-kitti-abs-rel-delta-1-25.md>): Video Depth Estimation; KITTI; Abs Rel: 0.076.

- [3D Reconstruction on Neural RGB-D (NRGBD)](<https://www.sota2.com/research/sota/3d-reconstruction-on-neural-rgb-d-nrgbd.md>): 3D Reconstruction; Neural RGB-D (NRGBD); Acc Mean: 0.055.

- [3D Reconstruction on 7-Scenes](<https://www.sota2.com/research/sota/3d-reconstruction-on-7-scenes.md>): 3D Reconstruction; 7 Scenes; Completion: not recorded.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2601.05208>)

[PDF](<https://arxiv.org/pdf/2601.05208.pdf>)
