Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Improving Diversity in Zero-Shot GAN Adaptation with Semantic Variations
Seogkyu Jeon, Bei Liu, Pilhyeon Lee, Kibeom Hong, Jianlong Fu, Hyeran Byun
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 |
|---|---|---|---|
| replace directional loss with directional moment loss over learned semantic variations (ablation) | zero-shot GAN adaptation baseline · AFHQ Dog-to-Cat | intra-cluster LPIPS (Avg.): 0.402 -> 0.464 | Improved |
| add elastic weight consolidation loss (ablation) | Baseline + L_dm · AFHQ Dog-to-Cat | intra-cluster LPIPS (Avg.): 0.464 -> 0.493 | Improved |
| add relation consistency loss (ablation) | Baseline + L_dm + L_EWC · AFHQ Dog-to-Cat | intra-cluster LPIPS (Avg.): 0.493 -> 0.507 | Improved |
| use L_EWC instead of layer selection in StyleGAN-NADA (ablation) | StyleGAN-NADA · AFHQ Dog-to-Cat | intra-cluster LPIPS (Avg.): 0.460 -> 0.480 | Improved |
| full method vs StyleGAN-NADA + L_EWC (ablation) | StyleGAN-NADA + L_EWC · AFHQ Dog-to-Cat | intra-cluster LPIPS (Avg.): 0.480 -> 0.507 | Improved |
| proposed zero-shot method vs 10-shot Ojha et al. on diversity | Ours · AFHQ Dog-to-Cat | intra-cluster LPIPS (Avg.): 0.575 -> 0.507 | Hurt |
| proposed method vs StyleGAN-NADA on FID in zero-shot | Ours · AFHQ Dog-to-Cat | FID: 70.225 -> 61.347 | Improved |
| extend method to 10-shot setting | Ours · AFHQ Dog-to-Cat 10-shot | FID: 57.196 -> 45.084 (vs Ojha et al.) | Improved |
| increase lambda_EWC beyond or below 10^7 (ablation) | Ours · AFHQ Dog-to-Cat | FID: 61.347 -> 94.317 (1e5), 81.758 (1e8) | Hurt |
| increase number of semantic variations K from 6 to 8 (ablation) | Ours · AFHQ Dog-to-Cat | FID: 61.347 -> 66.698 | Hurt |
Closest papers by meaning
- Adversarial Learning of Semantic Relevance in Text to Image Synthesis (AAAI 2019 · similarity 0.89)
- Generalized One-shot Domain Adaptation of Generative Adversarial Networks (NeurIPS 2022 · similarity 0.89)
- One-Shot Generative Domain Adaptation (ICCV 2023 · similarity 0.88)
- One-Shot Adaptation of GAN in Just One CLIP (TPAMI 2023 · similarity 0.88)
- On the Zero-shot Adversarial Robustness of Vision-Language Models: A Truly Zero-shot and Training-free Approach (CVPR 2025 · similarity 0.87)
- E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 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.
