Research map / Image generation / Editing and style transfer
Image Synthesis via Semantic Composition
Yi Wang, Lu Qi, Ying-Cong Chen, Xiangyu Zhang, Jiaya Jia
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 spatially conditional convolution (SCC) instead of standard conv, with BN (ablation) | SC-GAN · Cityscapes | FID: 69.7 -> 54.0 | Improved |
| use SCC instead of standard conv, with SCN (ablation) | SC-GAN · Cityscapes | FID: 60.9 -> 49.5 | Improved |
| use spatially conditional normalization (SCN) instead of BN, with standard conv (ablation) | SC-GAN · Cityscapes | FID: 69.7 -> 60.9 | Improved |
| use SCN instead of BN, with SCC (ablation) | SC-GAN · Cityscapes | FID: 54.0 -> 49.5 | Improved |
| add semantic vector generation loss Ls (L1) (ablation) | SC-GAN · Cityscapes | mIoU: 60.3 -> 66.9 | Improved |
| add semantic vector generation loss Ls (L1) (ablation) | SC-GAN · Cityscapes | FID: 82.5 -> 49.5 | Improved |
| use L2 norm instead of L1 for Ls (ablation) | SC-GAN · Cityscapes | mIoU: 66.9 -> 63.4 | Hurt |
| use VGG perceptual loss instead of L1 for Ls (ablation) | SC-GAN · Cityscapes | mIoU: 66.9 -> 64.2 | Hurt |
| increase weight candidates from n=1 to n=8 (ablation) | SCC · Cityscapes | mIoU: 61.2 -> 67.9 | Improved |
| apply softmax to semantic vectors instead of no nonlinearity (ablation) | SC-GAN · Cityscapes | FID: 73.9 -> 49.5 | Improved |
| use softmax instead of ReLU on semantic vectors (ablation) | SC-GAN · Cityscapes | FID: 66.5 -> 49.5 | Improved |
| use softmax instead of sigmoid on semantic vectors (ablation) | SC-GAN · Cityscapes | FID: 57.6 -> 49.5 | Improved |
Closest papers by meaning
- Semantic Image Synthesis With Spatially-Adaptive Normalization (CVPR 2019 · similarity 0.90)
- Photographic Image Synthesis With Cascaded Refinement Networks (ICCV 2017 · similarity 0.89)
- Efficient Semantic Image Synthesis via Class-Adaptive Normalization (TPAMI 2021 · similarity 0.88)
- SceneComposer: Any-Level Semantic Image Synthesis (CVPR 2023 · similarity 0.88)
- PLACE: Adaptive Layout-Semantic Fusion for Semantic Image Synthesis (CVPR 2024 · similarity 0.88)
- Unlocking Pre-trained Image Backbones for Semantic Image Synthesis (CVPR 2024 · similarity 0.86)
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.
