Research map / Detection and segmentation / Open-set and few-shot detection
Deep Spectral Methods: A Surprisingly Strong Baseline for Unsupervised Semantic Segmentation and Localization
Check 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 |
|---|---|---|---|
| use final attention key features rather than query, value, or output features (ablation) | DINO ViT-S localization · PASCAL VOC 2007 | CorLoc: 33.1 -> 61.6; 49.9 -> 61.6; 37.3 -> 61.6 | Helped |
| add color affinity to semantic features (ablation) | ViT localization · PASCAL VOC 2012 | CorLoc: 58.0 -> 61.9; 57.7 -> 61.2; 59.4 -> 62.6; 60.5 -> 62.7 | Helped |
| use color affinity alone instead of combining it with semantic features (ablation) | ViT-B-16 localization · PASCAL VOC 2012 | CorLoc: 57.7 -> 24.2 | Hurt |
| extract features from later ViT blocks (ablation) | DINO ViT-S-16 localization · PASCAL VOC 2007 | CorLoc: 28.2 -> 61.6 | Helped |
| use ConvNext or XCiT features instead of standard ViT features (ablation) | spectral localization · PASCAL VOC 2007 | CorLoc: 61.6 -> 41.8; 61.6 -> 33.6 | Hurt |
| use ConvNext features rather than ResNet-50 features (ablation) | spectral localization · PASCAL VOC 2007 | CorLoc: 26.6 -> 41.8 | Helped |
| use more than 3 eigenvectors for the initial semantic segmentation stage (ablation) | semantic segmentation pipeline · PASCAL VOC 2012 | mIoU: 29.7 -> 33.3 | Helped |
| vary the number of eigenvectors above 3 (ablation) | semantic segmentation pipeline · PASCAL VOC 2012 | mIoU: 33.3; 31.4; 31.8 | No clear effect |
| increase the number of semantic clusters K above 20 and use optimal matching (ablation) | semantic segmentation pipeline without self-training · PASCAL VOC 2012 | mIoU: 33.3 -> 38.5 -> 42.8 | Helped |
| use the proposed pseudo-label boxes for class-agnostic detection rather than LOST pseudo-label boxes | class-agnostic detection with self-training · PASCAL VOC 2007 | CorLoc: 64.5 -> 65.1 | Helped |
Closest papers by meaning
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 2024 · similarity 0.88)
- EAGLE: Eigen Aggregation Learning for Object-Centric Unsupervised Semantic Segmentation (CVPR 2024 · similarity 0.87)
- Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling (CVPR 2024 · similarity 0.87)
- Spectral Prompt Tuning: Unveiling Unseen Classes for Zero-Shot Semantic Segmentation (AAAI 2024 · similarity 0.87)
- Integrating Low-Level Visual Cues for Enhanced Unsupervised Semantic Segmentation (AAAI 2025 · similarity 0.86)
- Distilling Spectral Graph for Object-Context Aware Open-Vocabulary Semantic Segmentation (CVPR 2025 · 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.