Binarized Diffusion Model for Image Super-Resolution

Authors: Zheng Chen, Haotong Qin, Yong Guo, Xiongfei Su, Xin Yuan, Linghe Kong, Yulun Zhang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experiments demonstrate that our BI-Diff SR outperforms existing binarization methods.
Researcher Affiliation Academia 1Shanghai Jiao Tong University, 2ETH Zürich, 3Max Planck Institute for Informatics, 4Westlake University
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Code is released at: https://github.com/zhengchen1999/BI-Diff SR.
Open Datasets Yes We take DIV2K [59] and Flickr2K [33] as the training dataset.
Dataset Splits No The paper states training and testing datasets (DIV2K, Flickr2K for training; Manga109 for testing in ablation study), but does not explicitly define a separate validation dataset split with proportions or sample counts.
Hardware Specification Yes Our model is implemented based on Py Torch [47] with two Nvidia A100-80G GPUs.
Software Dependencies No The paper mentions PyTorch as the implementation framework but does not specify its version or any other software dependencies with version numbers.
Experiment Setup Yes For the noise estimation network, we set the encoder and decoder level to 4. ... We train models with the L1 loss. We employ the Adam optimizer [22] with β1=0.9 and β2=0.99, and a learning rate of 1 10 4. The batch size is set to 16, with a total of 1,000K iterations.