Research map / Image generation / Editing and style transfer
Semantic Image Synthesis with Unconditional Generator
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| replace LinearGAN latent optimization with feature-map rearranger, 1 mask | Ours · LSUN Church mIoU | mIoU: 16 -> 23.8 | Helped |
| same, 1 mask | Ours · LSUN Bedroom mIoU | mIoU: 17.5 -> 35.6 | Helped |
| same, 1 mask | Ours · FFHQ mIoU | mIoU: 37.2 -> 49.8 | Helped |
| increase number of masks from 1 to 16 | Ours · LSUN Church | mIoU: 23.8 -> 28.8 | Helped |
| increase number of masks from 1 to 16 | LinearGAN · LSUN Church | mIoU: 16 -> 20.4 | Helped |
| one-shot Ours instead of one-shot OASIS | Ours (1-shot) · CelebAMask-HQ | FID: 25.5 -> 18.5 | Helped |
| one-shot Ours instead of one-shot INADE | Ours (1-shot) · CelebAMask-HQ | mIoU: 44.3 -> 53.1 | Helped |
| one annotation instead of full 28K annotated pairs | Ours (1-shot) · CelebAMask-HQ | mIoU: 77.0 -> 53.1 | Hurt |
| one annotation instead of full 28K annotated pairs | Ours (1-shot) · CelebAMask-HQ | FID: 18.8 -> 18.5 | Mixed |
| add self-reconstruction loss to adversarial loss only (ablation) | Rearranger · CelebA | FID convergence speed | Helped |
| use L_self without L_mask (ablation) | Rearranger · CelebA | proxy mask adherence | Hurt |
| add mask reconstruction loss L_mask (ablation) | Rearranger · CelebA | proxy mask usage | Helped |
Closest papers by meaning
- Unlocking Pre-trained Image Backbones for Semantic Image Synthesis (CVPR 2024 · similarity 0.88)
- Learning to Predict Layout-to-image Conditional Convolutions for Semantic Image Synthesis (NeurIPS 2019 · similarity 0.88)
- MaskSketch: Unpaired Structure-Guided Masked Image Generation (CVPR 2023 · similarity 0.87)
- Diverse Semantic Image Synthesis via Probability Distribution Modeling (CVPR 2021 · similarity 0.87)
- PLACE: Adaptive Layout-Semantic Fusion for Semantic Image Synthesis (CVPR 2024 · similarity 0.85)
- Return of Unconditional Generation: A Self-supervised Representation Generation Method (NeurIPS 2024 · similarity 0.85)
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.