Research map / Image generation / Editing and style transfer
Editing Out-of-Domain GAN Inversion via Differential Activations
Haorui Song, Yong Du, Tianyi Xiang, Junyu Dong, Jing Qin, Shengfeng He
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 |
|---|---|---|---|
| add cam-consistency loss refinement to StarGAN | StarGAN · Celeba-HQ (4 attributes avg) | FID: 30.98 -> 28.52 | Improved |
| add cam-consistency loss refinement to AttGAN | AttGAN · Celeba-HQ (4 attributes avg) | FID: 17.96 -> 15.72 | Improved |
| add cam-consistency loss refinement to StarGAN | StarGAN · Celeba-HQ (4 attributes avg) | LPIPS: 0.208 -> 0.138 | Improved |
| add cam-consistency loss refinement to AttGAN | AttGAN · Celeba-HQ (4 attributes avg) | LPIPS: 0.107 -> 0.099 | Improved |
| use proposed Diff-CAM composition framework instead of best baseline AttGAN* | face attribute editing · Celeba-HQ (4 attributes avg) | FID: 15.72 -> 13.76 | Improved |
| use proposed framework instead of best baseline AttGAN* | face attribute editing · Celeba-HQ (4 attributes avg) | LPIPS: 0.099 -> 0.094 | Improved |
| replace Grad-CAM mask with differential activation (Diff-CAM) mask (ablation) | composition-decomposition editing · Celeba-HQ qualitative (age, eyes open) | mask coverage of edited region (qualitative) | Improved |
| add deghosting network after mask blending (ablation) | composition-decomposition editing · Celeba-HQ qualitative | ghosting artifacts (qualitative) | Improved |
| swap inversion encoder and direction method (pSp, IdInvert, e4e) (ablation) | proposed framework · Celeba-HQ qualitative | out-of-domain preservation (qualitative) | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- StarGAN — StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation (CVPR 2018) · better on 4 of 4 shared results
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
- In-Domain GAN Inversion for Faithful Reconstruction and Editability (TPAMI 2024 · similarity 0.94)
- In-Domain GAN Inversion for Real Image Editing (ECCV 2020 · similarity 0.93)
- Diverse Inpainting and Editing with GAN Inversion (ICCV 2023 · similarity 0.93)
- Gradual Residuals Alignment: A Dual-Stream Framework for GAN Inversion and Image Attribute Editing (AAAI 2024 · similarity 0.92)
- In-N-Out: Faithful 3D GAN Inversion with Volumetric Decomposition for Face Editing (CVPR 2024 · similarity 0.89)
- E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 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.
