Research map / Detection and segmentation / CNN features
DIH-CLIP: Unleashing the Diversity of Multi-Head Self-Attention for Training-Free Open-Vocabulary Semantic Segmentation
Songsong Duan, Xi Yang, Nannan 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 |
|---|---|---|---|
| add entropy-based head selection (ablation) | SCLIP · VOC21 | mIoU: 59.1 -> 60.6 | Improved |
| add entropy-based head selection (ablation) | SCLIP · City | mIoU: 32.2 -> 34.9 | Improved |
| add local-to-global transfer of head attention (ablation) | SCLIP · VOC21 | mIoU: 59.1 -> 59.9 | Improved |
| add local-to-global transfer of head attention (ablation) | SCLIP · Context60 | mIoU: 30.4 -> 31.1 | Improved |
| combine selection and transfer into SHA (ablation) | SCLIP · VOC21 | mIoU: 59.1 -> 62.3 | Improved |
| combine selection and transfer into SHA (ablation) | SCLIP · City | mIoU: 32.2 -> 36.7 | Improved |
| add middle-layer features (MLF) (ablation) | SCLIP · City | mIoU: 32.2 -> 35.3 | Improved |
| add middle-layer features on top of SHA (ablation) | SHA-SCLIP · VOC21 | mIoU: 62.3 -> 62.6 | Improved |
| full DIH-CLIP versus SHA with MLF (ablation) | SCLIP · VOC21 | mIoU: 62.6 -> 64.2 | Improved |
| plug SHA into ProxyCLIP (ablation) | ProxyCLIP · VOC21 | mIoU: 59.3 -> 61.2 | Improved |
| plug SHA into ProxyCLIP (ablation) | ProxyCLIP · City | mIoU: 37.0 -> 38.5 | Improved |
| plug SHA into NACLIP (ablation) | NACLIP · VOC21 | mIoU: 58.9 -> 61.9 | Improved |
Later papers that compared with this method
1 later accepted paper reports DIH-CLIP as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Looking Beyond the Window: Global-Local Aligned CLIP for Training-free Open-Vocabulary Semantic Segmentation (CVPR 2026 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- CLIP-DINOiser — CLIP-DINOiser: Teaching CLIP a few DINO tricks for open-vocabulary semantic segmentation (ECCV 2024) · better on 1 of 1 shared result
- ProxyCLIP — ProxyCLIP: Proxy Attention Improves CLIP for Open-Vocabulary Segmentation (ECCV 2024) · better on 1 of 1 shared result
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
- Enhancing Local Cognition of CLIP for Training-Free Open Vocabulary Semantic Segmentation (TPAMI 2026 · similarity 0.91)
- CLIPer: Hierarchically Improving Spatial Representation of CLIP for Open-Vocabulary Semantic Segmentation (ICCV 2025 · similarity 0.91)
- OPMapper: Enhancing Open-Vocabulary Semantic Segmentation with Multi-Guidance Information (NeurIPS 2025 · similarity 0.91)
- Emergent Open-Vocabulary Semantic Segmentation from Off-the-shelf Vision-Language Models (CVPR 2024 · similarity 0.90)
- Unveiling the Knowledge of CLIP for Training-Free Open-Vocabulary Semantic Segmentation (AAAI 2025 · similarity 0.90)
- ResCLIP: Residual Attention for Training-free Dense Vision-language Inference (CVPR 2025 · similarity 0.89)
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.
