Research map / Image generation / Editing and style transfer
Unsupervised Cross-Domain Image Generation
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 |
|---|---|---|---|
| Use DTN instead of the direct-input baseline | SVHN-to-MNIST transfer · SVHN to MNIST, transferred SVHN test samples | accuracy: 13.71% -> 90.66% | Helped |
| Add the target identity loss L_TID (ablation) | DTN · SVHN to MNIST, transferred SVHN test samples | accuracy: 88.40% -> 90.66% | Helped |
| Add the feature-constancy loss L_CONST (ablation) | DTN · SVHN to MNIST, transferred SVHN test samples | accuracy: 74.55% -> 90.66% | Helped |
| Include f as the input representation in G (ablation) | DTN · SVHN to MNIST, transferred SVHN test samples | accuracy: 36.90% -> 90.66% | Helped |
| Add the discriminator and GAN losses (ablation) | DTN · SVHN to MNIST, transferred SVHN test samples | accuracy: 34.70% -> 90.66% | Helped |
| Add either feature constancy or target identity regularization (ablation) | DTN · SVHN to MNIST, transferred SVHN test samples | accuracy: 5.28% -> 90.66% | Helped |
| Omit digit 3 from the source training set (ablation) | DTN · SVHN digit 3 to MNIST | accuracy: 94.67% -> 93.33% | No clear effect |
| Omit digit 3 from the target training set (ablation) | DTN · SVHN digit 3 to MNIST | accuracy: 94.67% -> 40.13% | Hurt |
| Omit digit 3 from both source and target training sets (ablation) | DTN · SVHN digit 3 to MNIST | accuracy: 94.67% -> 60.02% | Hurt |
| Omit digit 3 from source, target, and f training (ablation) | DTN · SVHN digit 3 to MNIST | accuracy: 94.67% -> 4.52 % | Hurt |
| Use DTN-generated emoji instead of manually created emoji | Face identity retrieval · Face-to-emoji retrieval, gallery of 100,001 face images | median rank: 16311 -> 16 | Helped |
| Replace image-distance L_TID with a second GAN (ablation) | DTN · Visual domain transfer | performance | No clear effect |
Closest papers by meaning
- Unsupervised Pixel-Level Domain Adaptation With Generative Adversarial Networks (CVPR 2017 · similarity 0.89)
- Few-Shot Image Generation via Cross-Domain Correspondence (CVPR 2021 · similarity 0.88)
- One-Shot Unsupervised Cross Domain Translation (NeurIPS 2018 · similarity 0.88)
- Unsupervised Domain Generalization by Learning a Bridge Across Domains (CVPR 2022 · similarity 0.88)
- Few-shot Cross-domain Image Generation via Inference-time Latent-code Learning (ICLR 2023 · similarity 0.88)
- Geometry-Consistent Generative Adversarial Networks for One-Sided Unsupervised Domain Mapping (CVPR 2019 · similarity 0.88)
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.