mlsanityData to 14 Oct 2026

A Principled Self-Referenced Early Stopping Approach for Deep Image Prior

arXiv 2605.25299 ↗First posted 24 May 2026cs.CV cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Recently, Deep Image Prior (DIP) has demonstrated strong capabilities for solving inverse imaging problems (IIPs) by optimizing a randomly initialized convolutional neural network in a training-data-free regime. However, DIP suffers from overfitting to noisy measurements due to network over-parameterization, making early stopping (ES) essential. The most successful ES method tracks fluctuations in the running variance of the network output to detect overfitting. However, in many applications, these fluctuations may appear prematurely, leading to unstable reconstructions. In this paper, we first show that nearly optimal DIP early stopping can be achieved when two independent noisy copies of the degraded image are available. Motivated by this observation, and since obtaining two fully independent copies is infeasible, we propose an overfitting detection framework based on constructing pseudo self-referenced images, resulting in three IIP-specific algorithms. Our approach is further supported by theoretical results on single-reference validation, pseudo-validation estimation, and the impact of shared noise. Across different IIPs, ranging from natural image restoration to medical image reconstruction, and under varying noise levels and noise types, our methods consistently outperform existing DIP early stopping approaches, all without requiring an accurate estimate of the noise level.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 20 of its 29 references that are in the map. First posted on arXiv 24 May 2026.

Builds on

20 of 29 references.

References · 20

  1. 2016Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoisingcited by 1,358
  2. 2017Deep Image Priorcited by 1,098
  3. 2021Advances in Neural Renderingcited by 365
  4. 2021NeRP: Implicit Neural Representation Learning with Prior Embedding for Sparsely Sampled Image Reconstructioncited by 146
  5. 2023Decomposed Diffusion Sampler for Accelerating Large-Scale Inverse Problemscited by 132
  6. 2018Deep Decoder: Concise Image Representations from Untrained Non-convolutional Networkscited by 120
  7. 2023Implicit Neural Representation in Medical Imaging: A Comparative Surveycited by 90
  8. 2017A Review of Convolutional Neural Networks for Inverse Problems in Imagingcited by 81
  9. 2018Image Restoration using Total Variation Regularized Deep Image Priorcited by 65
  10. 2019RARE: Image Reconstruction using Deep Priors Learned without Ground Truthcited by 64
  11. 2023Deep Learning on Implicit Neural Representations of Shapescited by 63
  12. 2024DMPlug: A Plug-in Method for Solving Inverse Problems with Diffusion Modelscited by 61
8 more
  1. 2021Early Stopping for Deep Image Priorcited by 44
  2. 2024Where Do We Stand with Implicit Neural Representations? A Technical and Performance Surveycited by 42
  3. 2021Recurrent Variational Network: A Deep Learning Inverse Problem Solver applied to the task of Accelerated MRI Reconstructioncited by 34
  4. 2022Gravitationally Lensed Black Hole Emission Tomographycited by 22
  5. 2025Diffusion models for inverse problemscited by 16
  6. 2024Learning Transferable Features for Implicit Neural Representationscited by 14
  7. 2024Analysis of Deep Image Prior and Exploiting Self-Guidance for Image Reconstructioncited by 11
  8. 2025A Baseline Method for Removing Invisible Image Watermarks using Deep Image Priorcited by 7

Built on by

Not cited yet.