PixelAsParam: A Gradient View on Diffusion Sampling with Guidance

Authors: Anh-Dung Dinh, Daochang Liu, Chang Xu

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The experimental results evidently improve over different baselines on datasets with various resolutions. (Abstract) and Section 6. Experiments
Researcher Affiliation Academia 1School of Computer Science, University of Sydney. Correspondence to: Anh-Dung Dinh <dinhanhdung1996@gmail.com>, Chang Xu <c.xu@sydney.edu.au>.
Pseudocode Yes Algorithm 1 DDPM denoising process with guidance and Algorithm 2 Pixel As Params Denoising Process (Px P)
Open Source Code Yes The code for the proposed sampling process is available here: https://github.com/dungdinhanh/pxpguided-diffusion
Open Datasets Yes All experiments are conducted on the Image Net dataset at 64x64, 128x128, 256x256, and CIFAR10 to evaluate our method since these datasets offer classification classes and pretrained diffusion models.
Dataset Splits No The paper uses well-known datasets but does not explicitly provide details about training, validation, or test dataset splits (e.g., percentages or sample counts).
Hardware Specification Yes All the experiments are run with cluster nodes with 8GPUs NVIDIA A-100.
Software Dependencies No The paper does not explicitly provide a list of software dependencies with specific version numbers (e.g., Python, PyTorch, or library versions).
Experiment Setup Yes All the hyperparameters for reproducing all the results are available in Table 7. (Table 7 provides specific values for hyperparameters like δ1, δ2, δ3, δ4, s, and TIME-STEPS for various models and datasets).