Research map / Detection and segmentation / Open-set and few-shot detection
CLIP-DINOiser: Teaching CLIP a few DINO tricks for open-vocabulary semantic segmentation
Monika Wysoczańska, Oriane Siméoni, Michaël Ramamonjisoa, Andrei Bursuc, Tomasz Trzciński, Patrick Pérez
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 |
|---|---|---|---|
| pool MaskCLIP features with DINO-based correlations instead of using MaskCLIP alone (ablation) | MaskCLIP · PASCAL VOC | mIoU: 32.9 -> 53.7 | Improved |
| learn pooling weights from CLIP features instead of using DINO weights (ablation) | CLIP-DINOiser · PASCAL VOC20 | mIoU: 79.1 -> 80.9 | Improved |
| pool at feature level instead of refining predictions with CLIP keys (ablation) | MaskCLIP · ADE20K | mIoU: 14.9 -> 20.4 | Improved |
| use CLIP key embeddings for feature pooling instead of DINO correlation (ablation) | CLIP-DINOiser · ADE20K | mIoU: 20.4 -> 7.7 | Hurt |
| add FOUND background detection (ablation) | CLIP-DINOiser · PASCAL VOC | mIoU: 53.7 -> 60.1 | Improved |
| relax FOUND with an uncertainty condition for background detection (ablation) | CLIP-DINOiser · PASCAL VOC | mIoU: 60.1 -> 62.1 | Improved |
| train on 10k ImageNet images instead of 1k (ablation) | CLIP-DINOiser · ADE20K and others | mIoU: 20.0 -> 20.1 (ADE) | No clear effect |
| use DINO value embeddings instead of key embeddings for correlation (ablation) | CLIP-DINOiser · qualitative segmentation | segmentation quality | Improved |
| use CLIP-DINOiser instead of OVDiff on object-centric VOC20 | open-vocabulary segmentation · PASCAL VOC20 | mIoU: 81.7 -> 80.9 | Hurt |
Later papers that compared with this method
6 later accepted papers report CLIP-DINOiser as a baseline in their results tables; 6 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)
- FLOSS: Free Lunch in Open-vocabulary Semantic Segmentation (ICCV 2025 · better on 5 of 5 shared results)
- Distilling Spectral Graph for Object-Context Aware Open-Vocabulary Semantic Segmentation (CVPR 2025 · better on 2 of 2 shared results)
- DeCLIP: Decoupled Learning for Open-Vocabulary Dense Perception (CVPR 2025 · better on 1 of 1 shared result)
- DIH-CLIP: Unleashing the Diversity of Multi-Head Self-Attention for Training-Free Open-Vocabulary Semantic Segmentation (ICCV 2025 · better on 1 of 1 shared result)
- 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
- Talking to DINO: Bridging Self-Supervised Vision Backbones with Language for Open-Vocabulary Segmentation (ICCV 2025 · similarity 0.93)
- Enhancing Local Cognition of CLIP for Training-Free Open Vocabulary Semantic Segmentation (TPAMI 2026 · similarity 0.92)
- Explore the Potential of CLIP for Training-Free Open Vocabulary Semantic Segmentation (ECCV 2024 · similarity 0.92)
- Convolutions Die Hard: Open-Vocabulary Segmentation with Single Frozen Convolutional CLIP (NeurIPS 2023 · similarity 0.91)
- Unveiling the Knowledge of CLIP for Training-Free Open-Vocabulary Semantic Segmentation (AAAI 2025 · similarity 0.91)
- 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.
