Research map / Image generation / GANs
InterFaceGAN: Interpreting the Disentangled Face Representation Learned by GANs
Yujun Shen, Ceyuan Yang, Xiaoou Tang, Bolei Zhou
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use StyleGAN W space instead of Z space for boundary correlation (ablation) | InterFaceGAN on StyleGAN · FF-HQ boundary correlation (Age-Glasses) | cosine similarity: 0.72 -> 0.05 | Improved |
| add conditional manipulation (subspace projection) (ablation) | InterFaceGAN on PGGAN · PGGAN attribute editing | qualitative entanglement | Improved |
| use W space instead of Z space for re-scoring Age manipulation (ablation) | InterFaceGAN on StyleGAN · Re-scoring, Glasses score when editing Age | score change: 0.32 -> 0.20 | Improved |
| edit only layers 08-17 instead of early layers (ablation) | InterFaceGAN on StyleGAN W · Layer-wise Pose re-scoring | score change: 0.42 -> 0.00 | Hurt |
| edit only layers 02-03 for gender (ablation) | InterFaceGAN on StyleGAN W · Layer-wise Gender re-scoring | score change: 0.45 -> 0.44 | Mixed |
| use StyleGAN W space instead of PGGAN Z space for identity preservation (Gender) (ablation) | InterFaceGAN · Identity discrepancy (Gender) | identity discrepancy: 0.61 -> 0.40 | Improved |
| edit smile instead of gender | InterFaceGAN on PGGAN · Identity discrepancy | identity discrepancy: 0.61 -> 0.24 | Improved |
| use W space instead of Z space for linear separability (ablation) | InterFaceGAN on StyleGAN · boundary classification accuracy | accuracy | Improved |
| use optimization-based instead of encoder-based GAN inversion | InterFaceGAN real editing · inversion time and identity | time per image: 3 minutes vs 0.1 seconds | Hurt |
| move latent code far from the boundary | InterFaceGAN on PGGAN · gender editing | identity preservation | Hurt |
| train pix2pixHD on synthetic pairs for smile | pix2pixHD with InterFaceGAN data · Real face editing | qualitative | Hurt |
Later papers that compared with this method
16 later accepted papers report InterfaceGAN as a baseline in their results tables; 14 of them report a better number on most of the results they share with it.
- PreciseControl: Enhancing Text-To-Image Diffusion Models with Fine-Grained Attribute Control (ECCV 2024 · better on 4 of 8 shared results)
- Region-Native Visual Tokenization (ECCV 2024 · better on 1 of 1 shared result)
- GlassesGAN: Eyewear Personalization Using Synthetic Appearance Discovery and Targeted Subspace Modeling (CVPR 2023 · better on 6 of 6 shared results)
- Text-Guided Unsupervised Latent Transformation for Multi-Attribute Image Manipulation (CVPR 2023 · better on 5 of 5 shared results)
- Talk-to-Edit: Fine-Grained 2D and 3D Facial Editing via Dialog (TPAMI 2023 · better on 5 of 5 shared results)
- FEditNet: Few-Shot Editing of Latent Semantics in GAN Spaces (AAAI 2023 · better on 3 of 4 shared results)
- FaceDNeRF: Semantics-Driven Face Reconstruction, Prompt Editing and Relighting with Diffusion Models (NeurIPS 2023 · better on 3 of 3 shared results)
- Multi-Directional Subspace Editing in Style-Space (ICCV 2023 · better on 3 of 3 shared results)
8 more not listed.
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Interpreting the Latent Space of GANs for Semantic Face Editing (CVPR 2020 · similarity 0.97)
- Conceptual and Hierarchical Latent Space Decomposition for Face Editing (ICCV 2023 · similarity 0.91)
- Unsupervised Disentanglement of Linear-Encoded Facial Semantics (CVPR 2021 · similarity 0.91)
- MOST-GAN: 3D Morphable StyleGAN for Disentangled Face Image Manipulation (AAAI 2022 · similarity 0.91)
- One-Shot Neural Face Reenactment via Finding Directions in GAN’s Latent Space (IJCV 2024 · similarity 0.91)
- SDGAN: Disentangling Semantic Manipulation for Facial Attribute Editing (AAAI 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
