跳到正文
返回论文列表
cs.GR提交于 已译

速度空间 3D 资产编辑

Velocity-Space 3D Asset Editing

Hao Liu · Yuxuan Lin · Jingfeng Guo · Ruihang Chu · Junjie Wang · Ruotong Li · et al.

中文摘要

局部 3D 资产编辑要求修改目标区域,同时保留其余内容。现有方法依赖生成器外部机制实现局部性,例如手工 3D 掩码、事后体素合并或 2D 多视图反投影,均未介入 ODE 采样器内部的退化根源。为实现忠实的局部编辑,整流流生成器的速度场应在目标区域保持强度,并在保留内容上消失;但单一速度场难以兼顾两者,进而产生身份泄漏、缺少专用编辑放大通道,以及几何和材质阶段的身份拖拽三类问题。VS3D(Velocity-Space 3D Asset editing)是一种无需反演、无需训练且无需掩码的框架,通过比较冻结模型在源条件与目标条件下的响应来推导编辑和保留信号,并设置三个阶段专用模块。重建锚定源注入(RASI)利用源重建校准每个资产的逐步锚点,以吸收身份泄漏。部分均值引导(PMG)对比速度差的高质量与低质量子采样估计,仅在存在一致编辑的位置放大编辑信号。孪生一致性残差注入(TAR)允许采样器逐 token 决定几何和材质阶段应保留的内容。多个 3D 资产上的实验表明,VS3D 优于最先进的原生 3D 与 2D 反投影编辑器,验证了速度空间编辑在免训练 3D 资产编辑中的有效性。代码公开于 https://github.com/LHaoooo/VS3D。

关键要点

  1. 01现有局部编辑依赖掩码、体素合并或多视图反投影,未解决 ODE 采样器内部的速度场退化问题。
  2. 02VS3D 通过比较冻结模型的源条件与目标条件响应,推导编辑信号与保留信号。
  3. 03RASI 以源重建校准逐步资产锚点,抑制保留区域中的身份泄漏。
  4. 04PMG 对比速度差的子采样估计,在一致编辑区域放大信号而不扰动身份。
  5. 05TAR 在几何与材质阶段逐 token 注入一致性残差;实验优于原生 3D 和 2D 反投影方法。

解读

尚无解读。

原始英文摘要

arXiv:2605.07385v2 Announce Type: replace-cross Abstract: Local 3D asset editing requires modifying a target region while preserving the rest. Existing methods enforce locality through mechanisms external to the generator, such as manual 3D masks, post-hoc voxel merging, or 2D multi-view lifting. None intervene where the corruption originates: inside the ODE sampler. For faithful local editing, a rectified-flow generator's velocity field should be strong over the target region while vanishing on preserved content. Yet a single velocity field can hardly satisfy both requirements, leading to three problems: (i) identity leakage that keeps the edit signal non-zero on preserved regions; (ii) no dedicated edit-amplification channel, so strengthening the edit perturbs identity; and (iii) identity drag at the geometry and material stages, where a global condition pulls every token toward the target. We propose VS3D (Velocity-Space 3D Asset editing), an inversion-free, training-free, and mask-free framework that derives edit and preserve signals by comparing the frozen model's source- and target-conditioned responses and realizes this principle through three stage-specific modules. Reconstruction-Anchored Source Injection (RASI) absorbs identity leakage by turning the unconditional embedding into a per-step, asset-specific anchor calibrated through source reconstruction. Partial-Mean Guidance (PMG) amplifies the edit signal by contrasting high- and low-quality subsample estimates of the velocity difference, active only where a consistent edit exists. Twin-Agreement Residual Injection (TAR) lets the sampler decide token by token what to preserve at the geometry and material stages. Experiments on diverse 3D assets show that VS3D outperforms state-of-the-art native-3D and 2D-lifted editors, demonstrating the effectiveness of velocity-space editing for training-free 3D asset editing. Code is publicly available at https://github.com/LHaoooo/VS3D.

同方向论文 · cs.GR

查看全部 →