Research map / Label-efficient and robust learning / Recognition and long-tail
Matching Adversarial Networks
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 |
|---|---|---|---|
| Replace the conditional GAN loss with MatAN β MS | Cityscapes semantic segmentation with a ResNet-50 generator · Cityscapes validation | mIoU: 22.0 -> 66.8 | Helped |
| Use multi-scale patches in MatAN β (ablation) | Cityscapes semantic segmentation · Cityscapes validation | mIoU: 63.3 -> 66.8 | Helped |
| Add the absolute-value distance between siamese features (ablation) | MatAN β for Cityscapes segmentation · Cityscapes validation | mIoU: 21.3 -> 63.3 | Helped |
| Add perturbations to the siamese branches (ablation) | MatAN for Cityscapes segmentation · Cityscapes validation | mIoU: 6.0 -> 63.3 | Helped |
| Switch the perturbation configuration from β to γ (ablation) | MatAN for Cityscapes segmentation · Cityscapes validation | mIoU: 63.3 -> 63.5 | No clear effect |
| Perturb the generated output as in configuration δ (ablation) | MatAN for Cityscapes segmentation · Cityscapes validation | mIoU: 63.3 -> 60.2 | Hurt |
| Add cross-entropy to multi-scale MatAN β (ablation) | MatAN for Cityscapes segmentation · Cityscapes validation | mIoU: 66.8 -> 65.1 | Hurt |
| Apply perturbations to generated outputs as well as ground-truth branches (ablation) | MatAN trained with perturbed labels · Cityscapes validation with perturbed ground truth | mIoU: 25.9 -> 58.1 | Helped |
| Replace the U-Net conditional GAN loss with MatAN β MS | Cityscapes semantic segmentation with a U-Net generator · Cityscapes validation | mIoU: 21.5 -> 48.9 | Helped |
| Use MatAN rather than CGAN for road centerline prediction | TorontoCity road centerline extraction · TorontoCity validation | F1: 75.7 -> 92.5 | Helped |
| Use MatAN rather than DWT for building instance segmentation | TorontoCity instance segmentation · TorontoCity validation | Pr. @50%: 75.1 -> 82.6 | Helped |
Closest papers by meaning
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.89)
- SGAN: An Alternative Training of Generative Adversarial Networks (CVPR 2018 · similarity 0.89)
- Training Generative Adversarial Networks with Limited Data (NeurIPS 2020 · similarity 0.89)
- Rob-GAN: Generator, Discriminator, and Adversarial Attacker (CVPR 2019 · similarity 0.89)
- Mixed batches and symmetric discriminators for GAN training (ICML 2018 · similarity 0.88)
- Adaptive Weighted Discriminator for Training Generative Adversarial Networks (CVPR 2021 · 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.