Revisiting Depth Representations for Feed-Forward 3D Gaussian Splatting

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Shi, Duochao, Wang, Weijie, Chen, Donny Y., Zhang, Zeyu, Bian, Jia-Wang, Zhuang, Bohan, Shen, Chunhua
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908395256152064
author Shi, Duochao
Wang, Weijie
Chen, Donny Y.
Zhang, Zeyu
Bian, Jia-Wang
Zhuang, Bohan
Shen, Chunhua
author_facet Shi, Duochao
Wang, Weijie
Chen, Donny Y.
Zhang, Zeyu
Bian, Jia-Wang
Zhuang, Bohan
Shen, Chunhua
contents Depth maps are widely used in feed-forward 3D Gaussian Splatting (3DGS) pipelines by unprojecting them into 3D point clouds for novel view synthesis. This approach offers advantages such as efficient training, the use of known camera poses, and accurate geometry estimation. However, depth discontinuities at object boundaries often lead to fragmented or sparse point clouds, degrading rendering quality -- a well-known limitation of depth-based representations. To tackle this issue, we introduce PM-Loss, a novel regularization loss based on a pointmap predicted by a pre-trained transformer. Although the pointmap itself may be less accurate than the depth map, it effectively enforces geometric smoothness, especially around object boundaries. With the improved depth map, our method significantly improves the feed-forward 3DGS across various architectures and scenes, delivering consistently better rendering results. Our project page: https://aim-uofa.github.io/PMLoss
format Preprint
id arxiv_https___arxiv_org_abs_2506_05327
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Revisiting Depth Representations for Feed-Forward 3D Gaussian Splatting
Shi, Duochao
Wang, Weijie
Chen, Donny Y.
Zhang, Zeyu
Bian, Jia-Wang
Zhuang, Bohan
Shen, Chunhua
Computer Vision and Pattern Recognition
Depth maps are widely used in feed-forward 3D Gaussian Splatting (3DGS) pipelines by unprojecting them into 3D point clouds for novel view synthesis. This approach offers advantages such as efficient training, the use of known camera poses, and accurate geometry estimation. However, depth discontinuities at object boundaries often lead to fragmented or sparse point clouds, degrading rendering quality -- a well-known limitation of depth-based representations. To tackle this issue, we introduce PM-Loss, a novel regularization loss based on a pointmap predicted by a pre-trained transformer. Although the pointmap itself may be less accurate than the depth map, it effectively enforces geometric smoothness, especially around object boundaries. With the improved depth map, our method significantly improves the feed-forward 3DGS across various architectures and scenes, delivering consistently better rendering results. Our project page: https://aim-uofa.github.io/PMLoss
title Revisiting Depth Representations for Feed-Forward 3D Gaussian Splatting
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2506.05327