Research map / Label-efficient and robust learning / Domain adaptation and generalization
DeceptionNet: Network-Driven Domain Randomization
Sergey Zakharov, Wadim Kehl, Slobodan Ilic
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 |
|---|---|---|---|
| guide augmentation with the adversarial deception network instead of unguided random perturbation | DeceptionNet · MNIST to MNIST-M | accuracy: 83.1 -> 90.4 | Improved |
| guide augmentation with the deception network instead of unguided random perturbation | DeceptionNet · Synthetic to Real Cropped LineMOD | accuracy: 53.1 -> 95.8 | Improved |
| guide augmentation with the deception network instead of unguided random perturbation | DeceptionNet · SYNTHIA to Cityscapes | mIoU: 22.3 -> 27.5 | Improved |
| add background module (ablation) | DeceptionNet · MNIST to MNIST-M | accuracy: 56.6 -> 82.4 | Improved |
| add noise module to background module (ablation) | DeceptionNet · MNIST to MNIST-M | accuracy: 82.4 -> 86.5 | Improved |
| add distortion module to background and noise (ablation) | DeceptionNet · MNIST to MNIST-M | accuracy: 86.5 -> 90.4 | Improved |
| add noise module to background module (ablation) | DeceptionNet · Synthetic to Real Cropped LineMOD | accuracy and angle error: 74.8 -> 77.6 accuracy, 50.4 -> 52.8 angle error | Mixed |
| add distortion module to background and noise (ablation) | DeceptionNet · Synthetic to Real Cropped LineMOD | accuracy and angle error: 77.6 -> 78.7 accuracy, 52.8 -> 48.2 angle error | Improved |
| add light module (ablation) | DeceptionNet · Synthetic to Real Cropped LineMOD | accuracy and angle error: 78.7 -> 95.8 accuracy, 48.2 -> 51.9 angle error | Mixed |
| use RGB-D input instead of RGB only (ablation) | DeceptionNet · Synthetic to Real Cropped LineMOD | accuracy: 84.8 -> 95.8 | Improved |
| use RGB-D input instead of depth only (ablation) | DeceptionNet · Synthetic to Real Cropped LineMOD | accuracy: 73.3 -> 95.8 | Improved |
| use RGB-D input instead of RGB only (ablation) | DeceptionNet · Synthetic to Real Cropped LineMOD | mean angle error: 57.4 -> 51.9 | Improved |
Closest papers by meaning
- Automated Synthetic-to-Real Generalization (ICML 2020 · similarity 0.86)
- DomainDrop: Suppressing Domain-Sensitive Channels for Domain Generalization (ICCV 2023 · similarity 0.86)
- Domain Generalization via Shuffled Style Assembly for Face Anti-Spoofing (CVPR 2022 · similarity 0.85)
- Dual Masked Generative Adversarial Transformer for Unsupervised Domain Adaptation (ECCV 2026 · similarity 0.85)
- Test-Time Domain Generalization for Face Anti-Spoofing (CVPR 2024 · similarity 0.85)
- Gradient Alignment for Cross-Domain Face Anti-Spoofing (CVPR 2024 · similarity 0.84)
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.
