Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
HifiFace: 3D Shape and Semantic Prior Guided High Fidelity Face Swapping
Authors: Yuhan Wang, Xu Chen, Junwei Zhu, Wenqing Chu, Ying Tai, Chengjie Wang, Jilin Li, Yongjian Wu, Feiyue Huang, Rongrong Ji
IJCAI 2021 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 4 Experiments |
| Researcher Affiliation | Collaboration | 1Youtu Lab, Tencent 2Zhejiang University 3Media Analytics and Computing Lab, Department of Artificial Intelligence, School of Informatics, Xiamen University 4Institute of Artificial Intelligence, Xiamen University |
| Pseudocode | No | The paper describes methods and equations, but does not include a dedicated pseudocode or algorithm block. |
| Open Source Code | Yes | Code is available at: https://johann.wang/HifiFace. |
| Open Datasets | Yes | We choose VGGFace2 [Cao et al., 2018] and Asian-Celeb [Deep Glint, 2020] as the training set. |
| Dataset Splits | No | We choose VGGFace2 [Cao et al., 2018] and Asian-Celeb [Deep Glint, 2020] as the training set. For our model with resolution 256 (i.e., Ours-256), we remove images with either size smaller than 256 for better image quality. The ratio of training pairs with the same identity is 50%. The test set contains 10K swapped faces and 10K real faces from FF++ for each method. |
| Hardware Specification | Yes | The model is trained with 200K steps, using 4 V100 GPUs and 32 batch size. |
| Software Dependencies | No | The paper mentions various models and algorithms by their citations (e.g., ADAM [Kingma and Ba, 2014], res-blocks [He et al., 2016], 3D face reconstruction model [Deng et al., 2019]), but does not specify software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | ADAM [Kingma and Ba, 2014] is used with β1 = 0; β2 = 0.99 and learning rate = 0.0001. The model is trained with 200K steps, using 4 V100 GPUs and 32 batch size. ... Lsid = λshape Lshape + λid Lid, where λid = 5 and λshape = 0.5. ... Lreal = Ladv + λ0Lseg + λ1Lrec + λ2Lcyc + λ3Llpips, where λ0 = 100, λ1 = 20, λ2 = 1 and λ3 = 5. |