Enhancing Identity-Deformation Disentanglement in StyleGAN for One-Shot Face Video Re-Enactment
Authors: Qing Chang, Yao-Xiang Ding, Kun Zhou
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results demonstrate the superiority of our approach compared to state-of-the-art methods. |
| Researcher Affiliation | Academia | Qing Chang, Yao-Xiang Ding*, Kun Zhou State Key Lab of CAD&CG, Zhejiang University, Hangzhou, China EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes the methodology and training process in narrative text and mathematical formulas, but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code https://nickchang97.github.io/Vi VFace.github.io/ |
| Open Datasets | Yes | First, we pre-train the entire network on the Celeb VHQ dataset (Zhu et al. 2022), which consists of 35K diverse videos. After this pre-training stage, we fine-tune our network on HDTF dataset (Zhang et al. 2021). This dataset contains about 300 high-resolution videos with over 300 identities. ... to compare with them. Datasets. To compare with previous SOTA (Yin et al. 2022; Oorloff and Yacoob 2023), we adopt similar training settings. ... we train our model on Celeb (Nagrani, Chung, and Zisserman 2017) that offers about 100k videos to compare with them. |
| Dataset Splits | Yes | Thirty videos of them are split as test set and the remaining videos are used as training set. For videos in the training set, we randomly sample 50 frames of each video for training. As for videos in the test set, we choose the first 500 frames of each video for evaluation. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU models, CPU types, memory) used for running the experiments. It only mentions that 'Training details are provided in the appendix.' which is not part of the main text. |
| Software Dependencies | No | The paper mentions several models and frameworks like Style GAN2, Stable Diffusion, ResNet50-SE, FPN, ArcFace, and Hair CLIP, along with their respective citations. However, it does not specify software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions or specific library versions). |
| Experiment Setup | Yes | Unless otherwise specified, the models in this section are trained solely on HDTF (Zhang et al. 2021) with 40K iterations. |