From Posterior Sampling to Meaningful Diversity in Image Restoration

Authors: Noa Cohen, Hila Manor, Yuval Bahat, Tomer Michaeli

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive user studies to analyze the proposed techniques, and find the strategy of reducing similarity between outputs to be significantly favorable over posterior sampling.
Researcher Affiliation Academia Noa Cohen Technion Israel Institute of Technology noa.cohen@campus.technion.ac.il Hila Manor Technion Israel Institute of Technology hila.manor@campus.technion.ac.il Yuval Bahat Princeton University yuval.bahat@gmail.com Tomer Michaeli Technion Israel Institute of Technology tomer.m@ee.technion.ac.il
Pseudocode Yes Algorithm 1 Hierarchical exploration
Open Source Code Yes Code and examples are available on the project s webpage.
Open Datasets Yes We experiment with two image restoration tasks: inpainting and noisy 16 super-resolution with a bicubic downsampling kernel and a noise level of 0.05. We analyze them in two domains: face images from the Celeb AMask-HQ dataset (Lee et al., 2020) and natural images from the Part Imagenet dataset (He et al., 2022).
Dataset Splits No The paper uses pre-trained models (e.g., Re Paint, DDRM, DDNM, DPS) and does not specify training, validation, or test dataset splits for these models. It mentions selecting 50 random images for user studies and generating sets of 100 images from the pre-trained models, but these are not training/validation/test splits for model development.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU models, CPU types) used for running its experiments.
Software Dependencies No The paper mentions 'Py Torch Toolbox for Image Quality Assessment available at https://github.com/chaofengc/IQAPyTorch' but does not specify version numbers for general software dependencies like Python, PyTorch, or CUDA.
Experiment Setup Yes Tabs. 4 and 5 lists the guidance parameters used in all figures and tables presented in the paper.