# DAPS++: Rethinking Diffusion Inverse Problems with Decoupled Posterior Annealing Source: [DAPS++: Rethinking Diffusion Inverse Problems with Decoupled Posterior Annealing]() Collection index: [llms.txt]() Data updated: 2026-07-01T05:00:04.992Z Authors: Hao Chen, Renzheng Zhang, Scott S. Howard Published: 2025-11-21T08:28:36.000Z ## Abstract From a Bayesian perspective, score-based diffusion solves inverse problems through joint inference, embedding the likelihood with the prior to guide the sampling process. However, this formulation fails to explain its practical behavior: the prior offers limited guidance, while reconstruction is largely driven by the measurement-consistency term, leading to an inference process that is effectively decoupled from the diffusion dynamics. We show that the diffusion prior in these solvers functions primarily as a warm initializer that places estimates near the data manifold, while reconstruction is driven almost entirely by measurement consistency. Based on this observation, we introduce \\textbf{DAPS++}, which fully decouples diffusion-based initialization from likelihood-driven refinement, allowing the likelihood term to guide inference more directly while maintaining numerical stability and providing insight into why unified diffusion trajectories remain effective in practice. By requiring fewer function evaluations (NFEs) and measurement-optimization steps, \\textbf{DAPS++} achieves high computational efficiency and robust reconstruction performance across diverse image restoration tasks. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Gaussian Blur on ImageNet 100 images (val)](): Gaussian Blur; ImageNet 100 images (val); SSIM: 66.6. - [Inpainting on ImageNet 100 images (val)](): Inpainting; ImageNet 100 images (val); SSIM: 77.1. - [Image Inpainting on FFHQ 100 images (val)](): Inpainting; FFHQ 100 images (val); SSIM: 81.2. - [4x Super-Resolution on ImageNet 100 images (val)](): Super-Resolution (SR x4); ImageNet 100 images (val); SSIM: 0.661. - [Motion Blur on ImageNet 100 images (val)](): Motion Blur; ImageNet 100 images (val); SSIM: 76.3. - [Motion Blur Deblurring on FFHQ 100 images (val)](): Motion Blur Deblurring; FFHQ 100 images (val); SSIM: 82.9. - [Nonlinear Blur on ImageNet 100 images (val)](): Nonlinear Blur; ImageNet 100 images (val); SSIM: 71.8. - [High Dynamic Range Reconstruction on FFHQ 100 images (val)](): High Dynamic Range (HDR) Reconstruction; FFHQ 100 images (val); SSIM: 0.834. - [Nonlinear Blur Deblurring on FFHQ 100 images (val)](): Nonlinear Blur Deblurring; FFHQ 100 images (val); SSIM: 74.5. - [Gaussian Blur Deblurring on FFHQ 100 images (val)](): Gaussian Blur Deblurring; FFHQ 100 images (val); SSIM: 0.784. - [High Dynamic Range (HDR) on ImageNet 100 images (val)](): High Dynamic Range (HDR); ImageNet 100 images (val); SSIM: 80.7. ## Original sources and implementation [Original paper]() [PDF]()