Research map / Image generation / Editing and style transfer
Positional Normalization
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 |
|---|---|---|---|
| Add PONO-MS to CycleGAN | CycleGAN · Maps: Photo → Map | FID: 58.3 -> 53.2 | Helped |
| Add PONO-MS to CycleGAN | CycleGAN · Horse ↔ Zebra | FID: 86.3 -> 71.2; 155.9 -> 142.2 | Helped |
| Add PONO-MS to Pix2pix | Pix2pix · Maps: Map → Photo | LPIPS: 0.333 -> 0.333 | No clear effect |
| Add PONO-MS to Pix2pix | Pix2pix · Cityscapes: Semantic labels → Photo | FID: 71.24 -> 60.40 | Helped |
| Add PONO-MS to DRIT | DRIT · Portrait → Photo | FID: 131.2 -> 127.9 | Helped |
| Add PONO-MS to MUNIT | MUNIT · Portrait → Photo | FID: 220.1 -> 270.5 | Hurt |
| Add PONO-MS to DRIT | DRIT · Cat → Dog | FID: 45.8 -> 47.5 | Hurt |
| Add one Conv3x3-LN-ReLU layer before the decoder output (ablation) | MUNIT + PONO-MS · Cat → Dog | FID: 315.6 -> 80.4 | Helped |
| Add PONO normalization alongside Moment Shortcut (ablation) | CycleGAN + Moment Shortcut · Maps: Map → Photo | FID: 54.5 -> 52.8 | Helped |
| Use multi-channel instead of one-channel DMS predictions (ablation) | CycleGAN + PONO-DMS (one 3×3 conv) · Horse → Zebra | FID: 74.1 -> 65.7 | Helped |
| Use two 3×3 convolutions with multi-channel DMS predictions (ablation) | CycleGAN + PONO-MS · Maps: Photo → Map | FID: 53.2 -> 51.4 | Helped |
| Replace the original BN with PONO-MS alone (ablation) | CycleGAN · Maps: Map → Photo | FID: 52.81 -> 49.59 | Hurt |
Closest papers by meaning
- Local Context Normalization: Revisiting Local Normalization (CVPR 2020 · similarity 0.85)
- Kalman Normalization: Normalizing Internal Representations Across Network Layers (NeurIPS 2018 · similarity 0.85)
- Normalizing the Normalizers: Comparing and Extending Network Normalization Schemes (ICLR 2017 · similarity 0.85)
- Norm matters: efficient and accurate normalization schemes in deep networks (NeurIPS 2018 · similarity 0.85)
- On the Importance of Gaussianizing Representations (ICML 2025 · similarity 0.84)
- Towards Training Without Depth Limits: Batch Normalization Without Gradient Explosion (ICLR 2024 · similarity 0.84)
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.