Research map / Detection and segmentation / Open-set and few-shot detection
Target Refocusing via Attention Redistribution for Open-Vocabulary Semantic Segmentation: An Explainability Perspective
Jiahao Li, Yang Lu, Yachao Zhang, Yong Xie, Fangyong Wang, Yuan Xie, Yanyun Qu
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 |
|---|---|---|---|
| mask distraction tokens with -inf (ablation) | Attn_kk-proxy CLIP · 4-benchmark average | mIoU: 32.1 -> 1.4 | Hurt |
| clamp distraction token embeddings to tau (low-pass filtering) (ablation) | Attn_kk-proxy CLIP · 4-benchmark average | mIoU: 32.1 -> 4.2 | Hurt |
| replace distraction token embeddings with 3x3 mean of neighbors (ablation) | Attn_kk-proxy CLIP · 4-benchmark average | mIoU: 32.1 -> 34.2 | Improved |
| replace distraction token embeddings with 3x3 median of neighbors (ablation) | Attn_kk-proxy CLIP · 4-benchmark average | mIoU: 32.1 -> 33.6 | Improved |
| 3x3 mean filtering on 10 random tokens (ablation) | layer-averaged Attn_kk CLIP · 4-benchmark average | mIoU: 32.9 -> 31.6 | Hurt |
| apply the same filtering on localized distraction tokens instead of random tokens (ablation) | RF-CLIP · 4-benchmark average | mIoU: 31.6 -> 34.0 | Improved |
| add attention weight redistribution (ablation) | RF-CLIP (III) · 4-benchmark average | mIoU: 34.0 -> 35.0 | Improved |
| add embedding weight redistribution (ablation) | RF-CLIP (III) · 4-benchmark average | mIoU: 34.0 -> 35.7 | Improved |
| add defocus localization to direct freed attention to target regions (ablation) | RF-CLIP (VI) · 4-benchmark average | mIoU: 36.6 -> 38.2 | Improved |
| send redistribution budget to defocused tokens instead of all non-distraction tokens (ablation) | RF-CLIP · 4-benchmark average | mIoU: 36.6 -> 38.2 | Improved |
| send redistribution budget to [CLS] token instead of non-distraction tokens (ablation) | RF-CLIP · 4-benchmark average | mIoU: 36.6 -> 34.2 | Hurt |
| enlarge embedding redistribution neighborhood from 3x3 to 7x7 (ablation) | RF-CLIP · 4-benchmark average | mIoU: 38.2 -> 36.1 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ProxyCLIP — ProxyCLIP: Proxy Attention Improves CLIP for Open-Vocabulary Segmentation (ECCV 2024) · better on 4 of 4 shared results
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
- OPMapper: Enhancing Open-Vocabulary Semantic Segmentation with Multi-Guidance Information (NeurIPS 2025 · similarity 0.91)
- Plug-in Feedback Self-adaptive Attention in CLIP for Training-free Open-Vocabulary Segmentation (ICCV 2025 · similarity 0.91)
- Unveiling the Knowledge of CLIP for Training-Free Open-Vocabulary Semantic Segmentation (AAAI 2025 · similarity 0.90)
- Open-Vocabulary Segmentation with Semantic-Assisted Calibration (CVPR 2024 · similarity 0.88)
- Emergent Open-Vocabulary Semantic Segmentation from Off-the-shelf Vision-Language Models (CVPR 2024 · similarity 0.88)
- Understanding Fine-tuning CLIP for Open-vocabulary Semantic Segmentation in Hyperbolic Space (CVPR 2025 · similarity 0.88)
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.
