Learning High-Quality Initial Noise for Single-View Synthesis with Diffusion Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhang, Zhihao, Yang, Xuejun, Liu, Weihua, Shen, Mouquan
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912773138546688
author Zhang, Zhihao
Yang, Xuejun
Liu, Weihua
Shen, Mouquan
author_facet Zhang, Zhihao
Yang, Xuejun
Liu, Weihua
Shen, Mouquan
contents Single-view novel view synthesis (NVS) models based on diffusion models have recently attracted increasing attention, as they can generate a series of novel view images from a single image prompt and camera pose information as conditions. It has been observed that in diffusion models, certain high-quality initial noise patterns lead to better generation results than others. However, there remains a lack of dedicated learning frameworks that enable NVS models to learn such high-quality noise. To obtain high-quality initial noise from random Gaussian noise, we make the following contributions. First, we design a discretized Euler inversion method to inject image semantic information into random noise, thereby constructing paired datasets of random and high-quality noise. Second, we propose a learning framework based on an encoder-decoder network (EDN) that directly transforms random noise into high-quality noise. Experiments demonstrate that the proposed EDN can be seamlessly plugged into various NVS models, such as SV3D and MV-Adapter, achieving significant performance improvements across multiple datasets. Code is available at: https://github.com/zhihao0512/EDN.
format Preprint
id arxiv_https___arxiv_org_abs_2512_16219
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Learning High-Quality Initial Noise for Single-View Synthesis with Diffusion Models
Zhang, Zhihao
Yang, Xuejun
Liu, Weihua
Shen, Mouquan
Computer Vision and Pattern Recognition
Image and Video Processing
Single-view novel view synthesis (NVS) models based on diffusion models have recently attracted increasing attention, as they can generate a series of novel view images from a single image prompt and camera pose information as conditions. It has been observed that in diffusion models, certain high-quality initial noise patterns lead to better generation results than others. However, there remains a lack of dedicated learning frameworks that enable NVS models to learn such high-quality noise. To obtain high-quality initial noise from random Gaussian noise, we make the following contributions. First, we design a discretized Euler inversion method to inject image semantic information into random noise, thereby constructing paired datasets of random and high-quality noise. Second, we propose a learning framework based on an encoder-decoder network (EDN) that directly transforms random noise into high-quality noise. Experiments demonstrate that the proposed EDN can be seamlessly plugged into various NVS models, such as SV3D and MV-Adapter, achieving significant performance improvements across multiple datasets. Code is available at: https://github.com/zhihao0512/EDN.
title Learning High-Quality Initial Noise for Single-View Synthesis with Diffusion Models
topic Computer Vision and Pattern Recognition
Image and Video Processing
url https://arxiv.org/abs/2512.16219