Research map / Detection and segmentation / CNN features
Adversarial Structure Matching for Structured Prediction Tasks
Jyh-Jing Hwang, Tsung-Wei Ke, Jianbo Shi, Stella X. Yu
Check 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 pixel-wise IID loss with ASM | FCN · PASCAL VOC 2012 val | mIoU (%): 68.91 -> 72.14 | Improved |
| Replace pixel-wise IID loss with ASM | DeepLab · PASCAL VOC 2012 val | mIoU (%): 77.54 -> 78.05 | Improved |
| Replace GAN with ASM | PSPNet · PASCAL VOC 2012 val | mIoU (%): 80.74 -> 81.43 | Improved |
| Replace IID loss with ASM using Conv1-2 analyzer features (ablation) | U-Net · Weizmann horse val | mIoU (%): 77.28 -> 79.62 | Improved |
| Use low- to mid-level Conv1-2 features rather than Conv1-3 features (ablation) | FCN with ASM · PASCAL VOC 2012 val | mIoU (%): 69.81 -> 71.60 | Improved |
| Remove structure reconstruction regularization (ablation) | ASM with U-Net · Weizmann horse val | mIoU (%): 79.62 -> 77.83 | Hurt |
| Remove structure reconstruction regularization (ablation) | ASM with FCN · PASCAL VOC 2012 val | mIoU (%): 71.60 -> 72.14 | No clear effect |
| Remove adversarial structure matching loss (ablation) | ASM with U-Net · Weizmann horse val | mIoU (%): 79.62 -> 76.70 | Hurt |
| Remove adversarial structure matching loss (ablation) | ASM with FCN · PASCAL VOC 2012 val | mIoU (%): 71.60 -> 71.26 | Hurt |
| Add IID loss alongside ASM (ablation) | ASM with U-Net · Weizmann horse val | mIoU (%): 79.62 -> 78.34 | Hurt |
| Add IID loss alongside ASM (ablation) | ASM with FCN · PASCAL VOC 2012 val | mIoU (%): 71.60 -> 68.49 | Hurt |
| Replace pixel-wise L2 with ASM | U-Net depth estimator · 2D-3D-S test depth | rel: 0.267 -> 0.252 | Improved |
Closest papers by meaning
- Efficient and Consistent Adversarial Bipartite Matching (ICML 2018 · similarity 0.85)
- Structure-Guided Adversarial Training of Diffusion Models (CVPR 2024 · similarity 0.85)
- Structured Convolutions for Efficient Neural Network Design (NeurIPS 2020 · similarity 0.84)
- Structured Gradient-based Interpretations via Norm-Regularized Adversarial Training (CVPR 2024 · similarity 0.84)
- SMaRt: Improving GANs with Score Matching Regularity (ICML 2024 · similarity 0.84)
- ASAM: Boosting Segment Anything Model with Adversarial Tuning (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.
