Research map / Detection and segmentation / Weakly supervised detection and segmentation
iSeg: An Iterative Refinement-Based Framework for Training-Free Segmentation
Lin E. Sun, Jiale Cao, Jin Xie, Fahad Shahbaz Khan, Yanwei Pang
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 |
|---|---|---|---|
| add entropy-reduced self-attention (Ent-Self) (ablation) | iSeg baseline · WSSS PASCAL VOC | mIoU: 68.2 -> 72.0 | Improved |
| add category-enhanced cross-attention (Cat-Cross) on top of Ent-Self (ablation) | iSeg · WSSS PASCAL VOC | mIoU: 72.0 -> 75.2 | Improved |
| add Ent-Self (ablation) | iSeg baseline · open-vocabulary PASCAL VOC | mIoU: 63.7 -> 67.1 | Improved |
| add Ent-Self (ablation) | iSeg baseline · WSSS MS COCO | mIoU: 40.1 -> 42.5 | Improved |
| add Ent-Self as DiffSeg post-processing (ablation) | iSeg baseline · unsupervised Cityscapes | mIoU: 22.8 -> 25.0 | Improved |
| increase iterations from 1 to 10 (ablation) | iSeg · WSSS PASCAL VOC | mIoU: 71.0 -> 75.2 | Improved |
| increase iterations from 10 to 12 (ablation) | iSeg · WSSS PASCAL VOC | mIoU: 75.2 -> 74.9 | No clear effect |
| use updating factor 0.01 instead of 0 (ablation) | iSeg · WSSS PASCAL VOC | mIoU: 69.1 -> 75.2 | Improved |
| raise updating factor from 0.01 to 0.1 (ablation) | iSeg · WSSS PASCAL VOC | mIoU: 75.2 -> 74.1 | Hurt |
| fuse 16x16 with 32x32 cross-attention maps (ablation) | iSeg · WSSS PASCAL VOC | mIoU: 74.8 -> 75.2 | Improved |
| fuse earlier decoder self-attention layers (#-3) instead of last layer only (ablation) | iSeg · WSSS PASCAL VOC | mIoU: 75.2 -> 68.5 | Hurt |
| use time-step 100 instead of 1 (ablation) | iSeg · WSSS PASCAL VOC | mIoU: 73.2 -> 75.2 | Improved |
Closest papers by meaning
- Repurposing Stable Diffusion Attention for Training-Free Unsupervised Interactive Segmentation (CVPR 2025 · similarity 0.87)
- InvSeg: Test-Time Prompt Inversion for Semantic Segmentation (AAAI 2025 · similarity 0.86)
- Integrating Low-Level Visual Cues for Enhanced Unsupervised Semantic Segmentation (AAAI 2025 · similarity 0.84)
- VidSeg: Training-free Video Semantic Segmentation based on Diffusion Models (CVPR 2025 · similarity 0.84)
- Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2025 · similarity 0.84)
- Diffuse Attend and Segment: Unsupervised Zero-Shot Segmentation using Stable Diffusion (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.
