Research map / Label-efficient and robust learning / Domain adaptation and generalization
Unsupervised Pixel-Level Domain Adaptation With Generative Adversarial Networks
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 |
|---|---|---|---|
| Adapt source images to the target pixel domain with PixelDA | MNIST source images · MNIST to USPS | classification accuracy (%): 78.9 -> 95.9 | Helped |
| Adapt source images to the target pixel domain with PixelDA | MNIST source images · MNIST to MNIST-M | classification accuracy (%): 63.6 (56.6) -> 98.2 | Helped |
| Adapt rendered images to the target pixel domain with PixelDA | Synthetic Cropped LineMod · Synthetic Cropped LineMod to Cropped LineMod | mean angle error: $89.2^{\circ}$ -> $\textbf{23.5}^{\circ}$ | Helped |
| Add the task loss on source images as well as adapted images (ablation) | PixelDA · Synthetic Cropped LineMod to Cropped LineMod | classification accuracy standard deviation: 22.32 -> 2.04 | Helped |
| Add the task loss on source images as well as adapted images (ablation) | PixelDA · Synthetic Cropped LineMod to Cropped LineMod | mean angle error standard deviation: 17.48 -> 3.24 | Mixed |
| Add the content-similarity loss to the task losses (ablation) | PixelDA · Synthetic Cropped LineMod to Cropped LineMod | classification accuracy standard deviation: 2.04 -> 1.60 | Helped |
| Add the content-similarity loss to the task losses (ablation) | PixelDA · Synthetic Cropped LineMod to Cropped LineMod | mean angle error standard deviation: 3.24 -> 6.97 | Hurt |
| Use ImageNet backgrounds instead of black backgrounds in source training images (ablation) | PixelDA, RGB-only · Synthetic Cropped LineMod to Cropped LineMod | classification accuracy (%): $94.16\%$ -> $96.95\% | Helped |
| Use ImageNet backgrounds instead of black backgrounds in source training images (ablation) | PixelDA, RGB-only · Synthetic Cropped LineMod to Cropped LineMod | mean angle error: $55.74^{\circ}$ -> $36.79^{\circ}$ | Helped |
| Add 1,000 labeled target examples to PixelDA training | PixelDA · Synthetic Cropped LineMod to Cropped LineMod, semi-supervised | mean angle error: $23.5^{\circ}$ -> $13.31^{\circ}$ | Helped |
Closest papers by meaning
- Adversarial Discriminative Domain Adaptation (CVPR 2017 · similarity 0.93)
- Generate to Adapt: Aligning Domains Using Generative Adversarial Networks (CVPR 2018 · similarity 0.92)
- Adversarial Feature Augmentation for Unsupervised Domain Adaptation (CVPR 2018 · similarity 0.92)
- Duplex Generative Adversarial Network for Unsupervised Domain Adaptation (CVPR 2018 · similarity 0.91)
- Learning From Synthetic Data: Addressing Domain Shift for Semantic Segmentation (CVPR 2018 · similarity 0.91)
- CrDoCo: Pixel-Level Domain Transfer With Cross-Domain Consistency (CVPR 2019 · similarity 0.91)
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.