Research map / Detection and segmentation / Open-set and few-shot detection
ProxyCLIP: Proxy Attention Improves CLIP for Open-Vocabulary Segmentation
Mengcheng Lan, Chaofeng Chen, Yiping Ke, Xinjiang Wang, Litong Feng, Wayne Zhang
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 adaptive normalization and masking to proxy attention (ablation) | ProxyCLIP (CLIP-ViT-B/16, DINO B/8) · 8 benchmarks avg | mIoU: 25.0 -> 42.3 | Improved |
| add masking on top of normalization (ablation) | ProxyCLIP · 8 benchmarks avg | mIoU: 29.5 -> 42.3 | Improved |
| use VFM feature similarity as proxy attention instead of VFM's own attention scores (ablation) | ProxyCLIP · 8 benchmarks avg | mIoU: 24.1 -> 42.3 (DINO B/8) | Improved |
| use DINO with patch size 8 instead of 16 (ablation) | ProxyCLIP · 8 benchmarks avg (CLIP-ViT-B/16) | mIoU: 41.1 -> 42.3 | Improved |
| use larger OpenCLIP-ViT-H/14 backbone instead of CLIP-ViT-B/16 (ablation) | ProxyCLIP · 8 benchmarks avg | mIoU: 42.3 -> 44.4 | Improved |
| scale CLIP backbone from B/16 to L/14 or H/14 in training-free SCLIP | SCLIP · 8 benchmarks avg | mIoU: 38.2 -> 29.0 (L/14), 29.1 (H/14) | Hurt |
| swap VFM among DINO, DINOv2, SAM (ablation) | ProxyCLIP · 8 benchmarks avg (CLIP-ViT-B/16) | mIoU: 41.1 / 41.1 / 40.8 | No clear effect |
| use Stable Diffusion UNet features as VFM instead of DINO B/16 (ablation) | ProxyCLIP · 8 benchmarks avg (CLIP-ViT-B/16) | mIoU: 41.1 -> 41.8 | Improved |
| use fixed hard-mask threshold instead of adaptive normalization (ablation) | ProxyCLIP · COCO Stuff | mIoU: DINO best 25.8 at 0.4 vs 26.5 adaptive | Mixed |
| use VFM embeddings instead of CLIP q-q attention in PAM (ablation) | ProxyCLIP · 8 benchmarks avg (CLIP-ViT-B/16) | mIoU: 38.2 -> 42.3 | Improved |
| add DINO proxy attention to CLIP | CLIP-ViT-B/16 · 8 benchmarks avg | mIoU: 11.7 -> 42.3 | Improved |
Later papers that compared with this method
19 later accepted papers report ProxyCLIP as a baseline in their results tables; 19 of them report a better number on most of the results they share with it.
- The Power of Prior: Training-Free Open-Vocabulary Semantic Segmentation with LLaVA (CVPR 2026 · better on 3 of 5 shared results)
- Looking Beyond the Window: Global-Local Aligned CLIP for Training-free Open-Vocabulary Semantic Segmentation (CVPR 2026 · better on 5 of 5 shared results)
- S2C2Seg: Semantic-Spatial Consistency and Category Optimization for Open-Vocabulary Segmentation (CVPR 2026 · better on 5 of 5 shared results)
- Target Refocusing via Attention Redistribution for Open-Vocabulary Semantic Segmentation: An Explainability Perspective (AAAI 2026 · better on 4 of 4 shared results)
- TextRegion: Text-Aligned Region Tokens from Frozen Image-Text Models (NeurIPS 2026 · better on 2 of 2 shared results)
- Direct Segmentation without Logits Optimization for Training-Free Open-Vocabulary Semantic Segmentation (CVPR 2026 · better on 2 of 2 shared results)
- Training-Free Open-Vocabulary Camouflaged Object Segmentation via Fine-Grained Object Binding and Adaptive Hybrid Prompt (CVPR 2026 · better on 2 of 2 shared results)
- From Visual Primitives to Semantic Masks: Fine-Grained Visual-Linguistic Alignment for Open-Vocabulary Remote Sensing Image Segmentation (ECCV 2026 · better on 1 of 1 shared result)
11 more not listed.
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.
- CLIP-DINOiser — CLIP-DINOiser: Teaching CLIP a few DINO tricks for open-vocabulary semantic 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
- Understanding Fine-tuning CLIP for Open-vocabulary Semantic Segmentation in Hyperbolic Space (CVPR 2025 · similarity 0.90)
- Unveiling the Knowledge of CLIP for Training-Free Open-Vocabulary Semantic Segmentation (AAAI 2025 · similarity 0.90)
- Emergent Open-Vocabulary Semantic Segmentation from Off-the-shelf Vision-Language Models (CVPR 2024 · similarity 0.90)
- Towards Open-Vocabulary Semantic Segmentation Without Semantic Labels (NeurIPS 2024 · similarity 0.89)
- Open-Vocabulary Segmentation with Semantic-Assisted Calibration (CVPR 2024 · similarity 0.89)
- Open-Vocabulary Semantic Segmentation with Image Embedding Balancing (CVPR 2024 · 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.
