Research map / Vision-language and multimodal / CLIP and zero-shot
PhaseAlign: Complex Phase Alignment for Stable Open-Vocabulary Semantic Segmentation
Jiankang Wang, Dingding Jia, Shuopeng Zhou, Xuan Wang
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 |
|---|---|---|---|
| use phase-only alignment instead of cosine-only similarity (ablation) | PhaseAlign · Similarity ablation | mIoU: 30.1 -> 37.3 | Improved |
| add cosine and in-batch contrastive terms to phase similarity (hybrid Sf) (ablation) | PhaseAlign · Similarity ablation | mIoU: 37.3 -> 40.9 | Improved |
| use phase-only instead of inner-product similarity (ablation) | PhaseAlign · Similarity ablation | BFScore: 3.55 -> 8.55 | Improved |
| add attention-guided affinity modeling (AAM) to CPA (ablation) | PhaseAlign · ADE20K / Cityscapes / Stuff | mIoU: 16.5 / 35.4 / 24.3 to 18.2 / 38.2 / 26.1 | Improved |
| add spatial-aware cross-modal projection (SACP) to CPA (ablation) | PhaseAlign · ADE20K / Cityscapes / Stuff | mIoU: 19.4 / 37.3 / 27.3 | Improved |
| combine CPA, AAM and SACP (full model) (ablation) | PhaseAlign · ADE20K / Cityscapes / Stuff | mIoU: 22.1 / 40.9 / 29.9 | Improved |
| replace MLP projection with SACP (ablation) | PhaseAlign · ADE20K / Cityscapes / Stuff | mIoU: 16.5 / 35.4 / 24.3 to 22.1 / 40.9 / 29.9 | Improved |
| replace MLP+Conv projection with SACP (ablation) | PhaseAlign · ADE20K / Cityscapes / Stuff | mIoU: 18.2 / 38.2 / 26.1 to 22.1 / 40.9 / 29.9 | Improved |
| use PhaseAlign instead of FreeDA under matched DINOv2-reg backbone, data, no priors | DINOv2-reg backbone · ADE20K / Cityscapes / Stuff | mIoU: surpassing FreeDA by 1.1, 2.3, and 2.8 points | Improved |
| use phase-based or hybrid similarity under visual feature scaling 0.5 to 4.0 (ablation) | PhaseAlign · Scale sensitivity (Figure 5) | mIoU / BFScore | Improved |
Closest papers by meaning
- DINOde: Continuous Vision-Text Alignment for Open-Vocabulary Semantic Segmentation (ECCV 2026 · similarity 0.87)
- Open-Vocabulary Segmentation with Semantic-Assisted Calibration (CVPR 2024 · similarity 0.86)
- Image-to-Image Matching via Foundation Models: A New Perspective for Open-Vocabulary Semantic Segmentation (CVPR 2024 · similarity 0.86)
- A Simple Framework for Open-Vocabulary Zero-Shot Segmentation (ICLR 2025 · similarity 0.86)
- Linguistic-Aware Patch Slimming Framework for Fine-grained Cross-Modal Alignment (CVPR 2024 · similarity 0.85)
- Training-Free Open-Vocabulary Segmentation with Offline Diffusion-Augmented Prototype Generation (CVPR 2024 · similarity 0.85)
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.
