Research map / Label-efficient and robust learning / Self-supervised and contrastive
The Spatially-Correlative Loss for Various Image Translation Tasks
Check 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 the content loss with FSeSim or LSeSim | CycleGAN architecture · Cityscapes | pixAcc: 68.8 -> 73.2 | Helped |
| Replace the content loss with FSeSim or LSeSim | CycleGAN architecture · Horse→Zebra | FID: 45.5 -> 38.0 | Helped |
| Use learned self-similarity loss instead of the MUNIT content loss | MUNIT architecture · Winter→Summer | FID: 97.4 ± 2.2 -> 89.4 ± 1.9 | Helped |
| Use learned self-similarity loss instead of the MUNIT content loss | MUNIT architecture · Night→Day | FID: 267.1 ± 2.7 -> 224.9 ± 2.0 | Helped |
| Organize self-similarity at random positions into local patches (ablation) | Spatially-correlative loss · Horse→Zebra | FID: 70.1 -> 45.8 | Helped |
| Replace global self-similarity with local patches (ablation) | Spatially-correlative loss · Night→Day | FID: 173.2 ± 2.2 -> 231.3 ± 2.5 | Hurt |
| Use multiple feature layers instead of one (ablation) | Local patch SeSim · Horse→Zebra | FID: 45.8 -> 43.3 | Helped |
| Use multiple feature layers instead of one (ablation) | Local patch SeSim · Night→Day | FID: 231.3 ± 2.5 -> 229.4 ± 2.1 | Helped |
| Replace L1 distance with cosine distance (ablation) | Multi-layer Patch SeSim · Horse→Zebra | FID: 43.3 -> 40.4 | Helped |
| Replace L1 distance with cosine distance (ablation) | Multi-layer Patch SeSim · Night→Day | FID: 229.4 ± 2.1 -> 234.3 ± 2.8 | Hurt |
| Learn the self-similarity representation with contrastive learning (ablation) | Cosine SeSim · Night→Day | FID: 234.3 ± 2.8 -> 224.9 ± 2.0 | Helped |
| Replace L1 distance with cosine distance (ablation) | Multi-layer Patch SeSim · Night→Day | LPIPS: 0.177 -> 0.203 | Helped |
Closest papers by meaning
- Towards Instance-Level Image-To-Image Translation (CVPR 2019 · similarity 0.90)
- Maximum Spatial Perturbation Consistency for Unpaired Image-to-Image Translation (CVPR 2022 · similarity 0.88)
- Alleviating Semantics Distortion in Unsupervised Low-Level Image-to-Image Translation via Structure Consistency Constraint (CVPR 2022 · similarity 0.88)
- Spatial Transport Optimization by Repositioning Attention Map for Training-Free Text-to-Image Synthesis (CVPR 2025 · similarity 0.86)
- Refining CLIP's Spatial Awareness: A Visual-Centric Perspective (ICLR 2025 · similarity 0.86)
- ENAT: Rethinking Spatial-temporal Interactions in Token-based Image Synthesis (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.