Identity-preserving Distillation Sampling by Fixed-Point Iterator

SeonHwa Kim1, Jiwon Kim1, Soobin Park2, Donghoon Ahn1, Jiwon Kang1,
Seungryong Kim3*, Kyong Hwan Jin1*, Eunju Cha2*
Korea University1, Sookmyung Women's University2, KAIST3
CVPR 2025
*Indicates Corresponding Authors
Code arXiv

Qualitative Results on IP2P datasets

Qualitative Results of NeRF on LLFF dataset

Source prompt: The green leaves. \( \rightarrow \) Target prompt: Yellow and red leaves in autumn.

Accumulation error in DDS

In DDS, misalignment between random noise \( \epsilon \) and estimated score \( \epsilon_{\phi}^{\text{src}} \) causes error.
This error accumulates during the optimization process, resulting in undesired changes such as structure and pose.

IDS Pipeline

To reduce the error caused by misalignment, we propose Identity-preserving Distillation Sampling (IDS) method by Fixed-Point Regularization (FPR). Updating the noisy source latent \( \mathbf{z}_{t}^{\text{src}} \) with FPR loss, IDS provides an edited result that is aligned with target prompt \( y^{\text{trg}} \) while maintaining the source's identity.

Abstract

Score distillation sampling (SDS) demonstrates a powerful capability for text-conditioned 2D image and 3D object generation by distilling the knowledge from learned score functions. However, SDS often suffers from blurriness caused by noisy gradients. When SDS meets the image editing, such degradations can be reduced by adjusting bias shifts using reference pairs, but the de-biasing techniques are still corrupted by erroneous gradients. To this end, we introduce Identity-preserving Distillation Sampling (IDS), which compensates for the gradient leading to undesired changes in the results. Based on the analysis that these errors come from the text-conditioned scores, a new regularization technique, called fixed-point iterative regularization (FPR), is proposed to modify the score itself, driving the preservation of the identity even including poses and structures. Thanks to a self-correction by FPR, the proposed method provides clear and unambiguous representations corresponding to the given prompts in image-to-image editing and editable neural radiance field (NeRF). The structural consistency between the source and the edited data is obviously maintained compared to other state-of-the-art methods.

Overall flowchart of IDS

MY ALT TEXT

BibTeX


@article{kim2025identity,
  title={Identity-preserving Distillation Sampling by Fixed-Point Iterator},
  author={Kim, SeonHwa and Kim, Jiwon and Park, Soobin and Ahn, Donghoon and Kang, Jiwon and Kim, Seungryong and Jin, Kyong Hwan and Cha, Eunju},
  journal={arXiv preprint arXiv:2502.19930},
  year={2025}
}