Research map / Detection and segmentation / Open-set and few-shot detection
Learning Neural Eigenfunctions for Unsupervised Semantic Segmentation
Zhijie Deng, Yucen Luo
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 |
|---|---|---|---|
| learn neural eigenfunctions for spectral clustering on ViT features instead of K-means | ViT-S features · Pascal Context val | mIoU: 28.9 -> 38.8 | Improved |
| use larger pre-trained ViT backbone (ablation) | Ours · Pascal Context val | mIoU: 38.8 (ViT-S) -> 33.2 (ViT-L) | Hurt |
| use features before the linear head (Ours*) with K-means instead of eigenfunction outputs (ablation) | Ours · Pascal Context val | mIoU: 38.8 -> 39.6 | Improved |
| output dimension K increased from 64 to 256 (ablation) | Ours · Pascal Context val | mIoU: 27.8 -> 38.8 | Improved |
| increase K from 256 to 512 (ablation) | Ours · Pascal Context val | mIoU: 38.8 -> 37.9 | No clear effect |
| add pixel-based graph (alpha=0.3) to feature graph (ablation) | Ours · Pascal Context val | mIoU: 38.6 -> 38.8 | No clear effect |
| add pixel-based graph (alpha=0.3) to feature graph (ablation) | Ours · Cityscapes val | mIoU: 27.2 -> 28.2 | Improved |
| vary beta from 0.04 to 0.16 (ablation) | Ours · Pascal Context val | mIoU: 38.8 -> 39.0 | No clear effect |
| train on ImageNet and test zero-shot on target datasets (ablation) | Ours · Pascal Context val | mIoU: 38.8 -> 15.2 | Hurt |
| apply method to ADE20K with 150 classes | Ours · ADE20K | mIoU: 19.2 -> 21.6 | Improved |
Closest papers by meaning
- Deep Spectral Methods: A Surprisingly Strong Baseline for Unsupervised Semantic Segmentation and Localization (CVPR 2022 · similarity 0.92)
- SpectralNet: Spectral Clustering using Deep Neural Networks (ICLR 2018 · similarity 0.90)
- Delving into Spectral Clustering with Vision-Language Representations (ICLR 2026 · similarity 0.89)
- Deciphering 'What' and 'Where' Visual Pathways from Spectral Clustering of Layer-Distributed Neural Representations (CVPR 2024 · similarity 0.87)
- EAGLE: Eigen Aggregation Learning for Object-Centric Unsupervised Semantic Segmentation (CVPR 2024 · similarity 0.86)
- Label-Efficient Few-Shot Semantic Segmentation with Unsupervised Meta-Training (AAAI 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.
