Diffusion Priors for Variational Likelihood Estimation and Image Denoising

Authors: Jun Cheng, Shan Tan

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments and analyses on diverse real-world datasets demonstrate the effectiveness of our method.
Researcher Affiliation Academia Jun Cheng, Shan Tan School of Artificial Intelligence and Automation, Huazhong University of Science and Technology jcheng24@hust.edu.cn, shantan@hust.edu.cn
Pseudocode Yes Algorithm 1 Difusion priors-based variational image denoising
Open Source Code Yes Code is available at https://github.com/HUST-Tan/Diffusion VI.
Open Datasets Yes We consider several real-world denoising datasets to evaluate our method, including SIDD [1], Poly U [47], CC [30], and FMDD [51].
Dataset Splits No The paper mentions 'SIDD validation' and dataset sizes, but does not specify explicit training/validation/test splits (e.g., percentages or sample counts) for all datasets used.
Hardware Specification Yes All experiments are conducted on Nvidia 2080Ti GPU.
Software Dependencies No The paper mentions using a pre-trained diffusion model but does not specify versions for core software libraries or dependencies like Python, PyTorch, or CUDA.
Experiment Setup Yes The total diffusion steps are 1000 by default, i.e., t [1, , 1000]. We choose α = 1 and Gaussian kernel size l = 9. The hyperparameters β and s for different datasets are summarized in Table 1. Different α/β represent the rough estimation of the prior precision for noises in different datasets, and Gaussian kernel scale s controls the range of local spatial correlation. The temperature γ is set to 1/5 for all datasets and will be ablated in the sequel.