Source-Agnostic Image Translation Based on Latent Aware Adaptive Masking

2026-08-14

AI Summary

This study addresses the challenges of source domain dependency and structural distortion caused by fixed masks in cross-domain image translation. We propose a source-independent adaptive masking framework that innovatively integrates a temporal adaptive statistical thresholding mechanism based on prediction discrepancies with latent space awareness and pretrained diffusion models to achieve dynamic, precise mask generation. Experimental results demonstrate that this approach effectively overcomes the limitations of conventional methods. On benchmarks such as AFHQ, our framework significantly outperforms existing unsupervised techniques in both FID and SSIM metrics, yielding substantial improvements in translation quality and structural fidelity.

Abstract

In this work, we propose a source-agnostic framework that dynamically refines a binary mask throughout the reverse diffusion process by computing the discrepancies of a pretrained diffusion model's prediction for each latent time step. Rather than relying on a fixed threshold, our method introduces a time-dependent statistical thresholding scheme derived from the empirical mean and standard deviation of prediction discrepancies across the latent noisy images from the target distribution. This allows the mask to adapt to the model's varying predictive confidence at different noise levels, effectively isolating domain-specific regions while preserving global structural coherence. Experimental results on the AFHQ and Celeba-HQ datasets demonstrate that our approach outperforms state-of-the-art unsupervised Image-to-Image methods in both realism (FID, KID) and faithfulness (SSIM, LPIPS). By requiring only a pretrained model of the target domain, our approach enables precise, automated localization and seamless translation across diverse source distributions without any specialized training. The project source code is available at: https://github.com/dtoma95/PM-Edit

Problem

• Source-Agnostic Image Translation
• Unsupervised Image-to-Image
• Domain Adaptation
• Mask Refinement

Innovation

• Source-Agnostic Image Translation
• Latent Aware Adaptive Masking
• Time-Dependent Statistical Thresholding
• Reverse Diffusion Process
• Unsupervised Image-to-Image