Efficient Conditional Diffusion Model with Probability Flow Sampling for Image Super-resolution

Authors: Yutao Yuan, Chun Yuan

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on DIV2K, Image Net, and Celeb A demonstrate that our method achieves higher super-resolution quality than existing diffusion-based image super-resolution methods while having lower time consumption.
Researcher Affiliation Academia Yutao Yuan, Chun Yuan Tsinghua University yuanyt21@mails.tsinghua.edu.cn, yuanc@sz.tsinghua.edu.cn
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks clearly labeled as such.
Open Source Code Yes Our code is available at https://github.com/Yuan-Yutao/ECDP.
Open Datasets Yes For general image super-resolution (4 ), we evaluate the performance of various methods on two datasets, DIV2K (Agustsson and Timofte 2017) and Image Net (Russakovsky et al. 2015). ... For face image super-resolution (8 ), we train and evaluate on Celeb A.
Dataset Splits No The paper describes data usage for training and evaluation (e.g., 'trained using random HR image crops', 'evaluated with full-size images', 'center cropped and resized'), but does not explicitly provide quantitative train/validation/test dataset splits (e.g., percentages or exact counts) or reference to standard numerical splits.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, memory, or cloud instance types) used for running the experiments.
Software Dependencies No The paper mentions software components and architectures like 'U-Net architecture', 'Big GAN residual blocks', 'VGG network', and 'Runge Kutta ODE solver', but does not provide specific version numbers for any of these or other software dependencies.
Experiment Setup Yes In our experiments, we set β(t) in our forward process (5) to a linear function increasing from β(0) = 0.1 and β(T) = 20, matching the settings in VP-SDE (Song et al. 2021). ... To generate HR images, we perform probability flow sampling using a standard Runge Kutta ODE solver with absolute and relative error tolerance of 10^-4.