Research map / Image generation / Editing and style transfer
ChunkyGAN: Real Image Inversion via Segments
Adéla Šubrtová, David Futschik, Jan Čech, Michal Lukáč, Eli Shechtman, Daniel Sýkora
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 |
|---|---|---|---|
| optimize separate latent codes per segment instead of one global code | W+ projection · CelebA 100 images, projection fidelity | LPIPS: 0.3675 -> 0.3194 | Improved |
| optimize separate latent codes per segment instead of one global code | W projection · CelebA 100 images, projection fidelity | LPIPS: 0.4190 -> 0.3697 | Improved |
| optimize separate latent codes per segment instead of one global code | S-space projection · CelebA 100 images, projection fidelity | LPIPS: 0.3577 -> 0.3572 | No clear effect |
| optimize separate latent codes per segment | W+ projection · CelebA 100 images, projection fidelity | L2 loss: 0.0436 -> 0.0207 | Improved |
| optimize separate latent codes per segment | W+ projection · CelebA 100 images, projection fidelity | identity loss: 0.1195 -> 0.0937 | Improved |
| add latent code regularization lambda_reg=1 (ablation) | Ours in W+ · CelebA 100 images, projection fidelity | LPIPS: 0.3194 -> 0.3330 | Hurt |
| add latent code regularization lambda_reg=1 (ablation) | Ours in W+ · CelebA 100 images, projection fidelity | identity loss: 0.0937 -> 0.0894 | Improved |
| add latent code regularization lambda_reg=1 (ablation) | Ours in W+ · CelebA editing, gender, projected vs edited | identity loss: 0.298 -> 0.126 | Improved |
| add latent code regularization lambda_reg=1 (ablation) | Ours in W+ · CelebA editing, beard, projected vs edited | identity loss: 0.312 -> 0.091 | Improved |
| add latent code regularization lambda_reg=1 (ablation) | Ours in W+ · CelebA editing, age, projected vs edited | identity loss: 0.151 -> 0.069 | Improved |
| use segment-wise codes without regularization for editing | Ours in W+ · CelebA editing, gender, projected vs edited | identity loss: 0.209 -> 0.298 | Hurt |
Closest papers by meaning
- EditGAN: High-Precision Semantic Image Editing (NeurIPS 2021 · similarity 0.89)
- ReGANIE: Rectifying GAN Inversion Errors for Accurate Real Image Editing (AAAI 2023 · similarity 0.89)
- High-Fidelity GAN Inversion for Image Attribute Editing (CVPR 2022 · similarity 0.88)
- In-Domain GAN Inversion for Faithful Reconstruction and Editability (TPAMI 2024 · similarity 0.88)
- Neural Photo Editing with Introspective Adversarial Networks (ICLR 2017 · similarity 0.88)
- In-Domain GAN Inversion for Real Image Editing (ECCV 2020 · similarity 0.88)
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.
