Research map / Detection and segmentation / Open-set and few-shot detection
ClearCLIP: Decomposing CLIP Representations for Dense Vision-Language Inference
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 |
|---|---|---|---|
| remove the residual connection in the last transformer block (ablation) | CLIP ViT-B/16 · 5 datasets without background | mIoU (Avg.): 27.3 -> 33.7 | Improved |
| remove the FFN alone in the last block (ablation) | CLIP ViT-B/16 · 5 datasets without background | mIoU (Avg.): 27.3 -> 25.9 | Hurt |
| remove both residual connection and FFN (ablation) | CLIP ViT-B/16 · 5 datasets without background | mIoU (Avg.): 33.7 -> 37.5 | Improved |
| use q-q self-self attention instead of q-k attention with residual and FFN removed (ablation) | ClearCLIP · 5 datasets without background | mIoU (Avg.): 27.6 -> 37.5 | Improved |
| q-q and k-k self-self attention on larger CLIP-L/14 without removing residual (ablation) | CLIP-L/14 · 5 datasets without background | mIoU | Hurt |
| increase scaling factor alpha of attention output over residual (ablation) | CLIP · 5 datasets without background | mIoU | Improved |
| ClearCLIP vs SCLIP on ViT-L/14 | CLIP ViT-L/14 · 5 datasets without background | mIoU (Avg.): 23.6 -> 34.5 | Improved |
| ClearCLIP vs SCLIP on ViT-B/16 with background class | CLIP ViT-B/16 · VOC21 | mIoU: SCLIP 59.1 vs ClearCLIP 51.8 (SCLIP* reproduced 51.4) | Mixed |
| ClearCLIP vs TCL (weakly supervised) with background class | CLIP ViT-B/16 · 3 datasets with background | mIoU (Avg.): 35.3 -> 39.1 | Improved |
| ClearCLIP on ViT-B/16 vs ViT-L/14 backbone | ClearCLIP · 5 datasets without background | mIoU (Avg.): 37.5 -> 34.5 | Hurt |
Later papers that compared with this method
13 later accepted papers report ClearCLIP as a baseline in their results tables; 13 of them report a better number on most of the results they share with it.
- CURE: Contextual Debiasing and Unbiased Refinement for Training-Free Open-Vocabulary Semantic Segmentation (ECCV 2026 · better on 6 of 7 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)
- ReAttnCLIP: Training-Free Open-Vocabulary Remote Sensing Image Segmentation via Re-defined Attention in CLIP (CVPR 2026 · better on 1 of 1 shared result)
- Looking Beyond the Window: Global-Local Aligned CLIP for Training-free Open-Vocabulary Semantic Segmentation (CVPR 2026 · better on 1 of 1 shared result)
- PEARL: Geometry Aligns Semantics for Training-Free Open-Vocabulary Semantic Segmentation (CVPR 2026 · better on 1 of 1 shared result)
- 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)
- SegEarth-OV: Towards Training-Free Open-Vocabulary Segmentation for Remote Sensing Images (CVPR 2025 · better on 9 of 9 shared results)
- Unveiling the Knowledge of CLIP for Training-Free Open-Vocabulary Semantic Segmentation (AAAI 2025 · better on 7 of 7 shared results)
5 more not listed.
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
- Unveiling the Knowledge of CLIP for Training-Free Open-Vocabulary Semantic Segmentation (AAAI 2025 · similarity 0.93)
- ResCLIP: Residual Attention for Training-free Dense Vision-language Inference (CVPR 2025 · similarity 0.92)
- Understanding Fine-tuning CLIP for Open-vocabulary Semantic Segmentation in Hyperbolic Space (CVPR 2025 · similarity 0.90)
- Emergent Open-Vocabulary Semantic Segmentation from Off-the-shelf Vision-Language Models (CVPR 2024 · similarity 0.90)
- Demystifying CLIP Data (ICLR 2024 · similarity 0.89)
- Towards Open-Vocabulary Semantic Segmentation Without Semantic Labels (NeurIPS 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.
