ACFun: Abstract-Concrete Fusion Facial Stylization

Authors: Jiapeng Ji, Kun Wei, Ziqi Zhang, Cheng Deng

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we conduct experiments using images collected from the Internet and provide visual comparisons.
Researcher Affiliation Academia Jiapeng Ji, Kun Wei , Ziqi Zhang, Cheng Deng School of Electronic Engineering, Xidian University Xi an 710071, China jiapengji777@gmail.com, weikunsk@gmail.com, zqzh9116@gmail.com, chdeng.xd@gmail.com
Pseudocode No The paper includes architectural diagrams (Figure 2 and Figure 3) and descriptive text for its components, but it does not contain any formal pseudocode blocks or algorithms labeled as such.
Open Source Code No The paper does not provide open access to the data and code due to proprietary restrictions. However, detailed descriptions of the experimental setup and model architecture are provided in Sections 3 and 4, ensuring the experiments can be understood and replicated by researchers with similar resources.
Open Datasets No We conduct experiments using images collected from the Internet and provide visual comparisons.
Dataset Splits No The paper mentions training on 'a single pair of images' and evaluates performance, but it does not specify explicit train/validation/test splits with percentages or sample counts for the datasets used in its experiments.
Hardware Specification Yes We trained on a single Nvidia A6000 graphics card, and in the case of a single pair of images, we set the batch size to 1.
Software Dependencies No The paper mentions using 'Stable Diffusion' as its backbone model (e.g., SD1.5, SDXL, SD1.4) and 'CLIP' for encoding, but it does not provide specific version numbers for these or any other software dependencies.
Experiment Setup Yes We set the base learning rate to 1.0e 04, and the remaining hyperparameters are consistent with Stable Diffusion without changing. Through 40 steps of diffusion, our method can obtain stylized facial images with good results. We set the hyperparameters γ and β to 0.8 and 1.0, respectively, and all subsequent experiments will use this hyperparameter setting method.