Search papers, labs, and topics across Lattice.
This paper introduces a non-parametric method for image inpainting that utilizes data-guided stochastic dynamics to recover missing regions without the need for network training. By employing a reverse-time stochastic differential equation with a kernel-weighted correction derived from a reference dataset, the method effectively guides pixel evolution towards high-density areas of the data distribution. Experimental results on datasets such as MNIST and CelebA reveal that this approach significantly outperforms traditional inpainting techniques in terms of PSNR, SSIM, and visual quality, highlighting the potential of empirical reference statistics in image reconstruction tasks.
Non-parametric image inpainting using stochastic dynamics outperforms traditional methods without requiring any neural network training.
Image inpainting aims to recover missing regions while preserving structural consistency. We propose a non-parametric method without network training based on data-guided stochastic dynamics. Starting from a masked image, the missing pixels are evolved through a reverse-time stochastic differential equation with a kernel-weighted correction estimated directly from a reference dataset. This empirical correction guides the reconstruction toward high-density regions of the data distribution without training a neural network or fitting a parametric density model. Experiments on MNIST, Fashion-MNIST, and MVTec show that the proposed method outperforms Mean Fill, Telea, and Navier-Stokes inpainting in PSNR, SSIM, and visual quality. On CelebA, it remains competitive and produces plausible completions for structure-sensitive occlusions. These results demonstrate the effectiveness of empirical reference statistics as a non-parametric prior for image inpainting.
