Research map / Label-efficient and robust learning / Domain adaptation and generalization
Few-Shot Semantic Image Synthesis With Class Affinity Transfer
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 |
|---|---|---|---|
| Initialize with combined class affinities rather than random affinities (ablation) | OASIS · COCO-Stuff→ADE20K | FID: 54.0 -> 40.9 | Helped |
| Initialize with combined class affinities rather than random affinities (ablation) | PITI · COCO-Stuff→ADE20K | mIoU: 11.6 -> 22.3 | Helped |
| Freeze early discriminator layers (ablation) | OASIS · COCO-Stuff→ADE20K | FID: 87.2 -> 65.7 | Helped |
| Use two-stage rather than one-stage generator finetuning (ablation) | OASIS · COCO-Stuff→ADE20K | FID: 65.7 -> 55.9 | Helped |
| Add residual convolutional branches (ablation) | OASIS · ADE20K→COCO-Stuff | FID: 83.4 -> 79.9 | Helped |
| Freeze the decoder during finetuning (ablation) | PITI · COCO-Stuff→ADE20K | FID: 56.5 -> 52.4 | Helped |
| Add trainable encoder prompts (ablation) | PITI · COCO-Stuff→ADE20K | FID: 52.4 -> 51.1 | Helped |
| Add the class affinity matrix (ablation) | PITI · COCO-Stuff→ADE20K | mIoU: 14.1 -> 22.3 | Helped |
| Use separate rather than shared scale and shift paths (ablation) | OASIS · COCO-Stuff→ADE20K | mIoU: 30.1 -> 31.4 | Helped |
| Binarize affinities rather than use soft affinity weights (ablation) | OASIS · COCO-Stuff→ADE20K | FID: 44.7 -> 40.9 | Helped |
| Binarize affinities rather than use soft affinity weights (ablation) | PITI · COCO-Stuff→ADE20K | mIoU: 19.8 -> 22.3 | Helped |
| Combine segmentation, self-supervised, and text affinity estimates by majority vote (ablation) | CAT-OASIS · COCO-Stuff→ADE20K and ADE20K→COCO-Stuff | FID: 42.0 -> 40.9; 58.2 -> 53.7 | Helped |
Closest papers by meaning
- You Only Need Adversarial Supervision for Semantic Image Synthesis (ICLR 2021 · similarity 0.91)
- Unlocking Pre-trained Image Backbones for Semantic Image Synthesis (CVPR 2024 · similarity 0.90)
- Learning to Predict Layout-to-image Conditional Convolutions for Semantic Image Synthesis (NeurIPS 2019 · similarity 0.90)
- Learning Pixel-Level Semantic Affinity With Image-Level Supervision for Weakly Supervised Semantic Segmentation (CVPR 2018 · similarity 0.89)
- Linear Semantics in Generative Adversarial Networks (CVPR 2021 · similarity 0.89)
- Iterative Self-Training with Class-Aware Text-to-Image Synthesis for Visual Task Learning (AAAI 2025 · similarity 0.87)
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.