Research map / Detection and segmentation / CNN features
Learning from Adversity: Semantic-Aware Mask Refinement through Adversarial Perturbation
Beom Young Kim, Sung Ju Hwang
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 |
|---|---|---|---|
| refine with Phoenix instead of SAMRefiner | PointWSSIS 1% masks · LVIS PointWSSIS 1% | AP mask: 21.8 -> 28.7 | Improved |
| refine with Phoenix | Mask R-CNN R50 · LVIS | AP mask: 39.8 -> 46.9 | Improved |
| refine strong models with Phoenix | MaskDINO · LVIS | AP mask: 56.8 -> 58.0 | Improved |
| refine with SAMRefiner on strong model | MaskDINO · LVIS | AP mask: 56.8 -> 57.0 | No clear effect |
| refine Cityscapes masks with SAMRefiner | Mask R-CNN R50 · Cityscapes | AP: 36.1 -> 32.5 | Hurt |
| refine DIS masks with Phoenix | DIS U-Net · DIS average | IoU: 54.7 -> 75.7 | Improved |
| add adversarial semantic-aware perturbation instead of morphological noise (ablation) | Phoenix · LVIS PointWSSIS 1% | AP mask: 23.8 -> 28.7 | Improved |
| add contrastive CMRL loss (ablation) | Phoenix · LVIS PointWSSIS 1% | AP mask: 26.9 -> 28.7 | Improved |
| apply adversarial perturbation to SegRefiner (ablation) | SegRefiner · DIS average (U-Net) | IoU: 58.7 -> 71.8 | Improved |
| train Phoenix on masks from real models instead of adversarial noise (ablation) | Phoenix · DIS average (U-Net) | IoU: 75.7 -> 67.6 | Hurt |
| higher IoU threshold tau (ablation) | Phoenix · LVIS PointWSSIS 1% | AP mask: 28.2 -> 24.2 | Mixed |
| change adversarial criterion (Dice vs L1) (ablation) | Phoenix · LVIS PointWSSIS 1% | AP mask: 27.7 -> 28.7 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- +SAMRefiner — SAMRefiner: Taming Segment Anything Model for Universal Mask Refinement (ICLR 2025) · better on 7 of 7 shared results
- MaskDINO — Mask DINO: Towards a Unified Transformer-Based Framework for Object Detection and Segmentation (CVPR 2023) · better on 1 of 1 shared result
- ISNet — ISNet: Integrate Image-Level and Semantic-Level Context for Semantic Segmentation (ICCV 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Evaluating the Adversarial Robustness of Semantic Segmentation: Trying Harder Pays Off (ECCV 2024 · similarity 0.89)
- Dynamic Divide-and-Conquer Adversarial Training for Robust Semantic Segmentation (ICCV 2021 · similarity 0.88)
- Calibrated Adversarial Refinement for Stochastic Semantic Segmentation (ICCV 2021 · similarity 0.88)
- An Adversarial Perturbation Oriented Domain Adaptation Approach for Semantic Segmentation (AAAI 2020 · similarity 0.87)
- Adversarially Masking Synthetic To Mimic Real: Adaptive Noise Injection for Point Cloud Segmentation Adaptation (CVPR 2023 · similarity 0.87)
- ASAM: Boosting Segment Anything Model with Adversarial Tuning (CVPR 2024 · similarity 0.86)
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.
