Research map / Detection and segmentation / Open-set and few-shot detection
DINOde: Continuous Vision-Text Alignment for Open-Vocabulary Semantic Segmentation
Sung-Hoon Yoon, Hoyong Kwon, Changgyoon Oh, KUK-JIN YOON
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 |
|---|---|---|---|
| replace MLP text-to-DINO alignment with ODE-based Semantic Text Flow (ablation) | DINOv3 ViT-L + CLIP ViT-L · OVSS 8-benchmark average | mIoU: 47.7 -> 48.2 | Improved |
| add Velocity Tangent Projection to STF (ablation) | DINOde · OVSS 8-benchmark average | mIoU: 48.2 -> 48.5 | Improved |
| add Global Context Flow on CLS token to STF (ablation) | DINOde · OVSS 8-benchmark average | mIoU: 48.2 -> 49.2 | Improved |
| add VTP on top of STF+GCF (ablation) | DINOde · OVSS 8-benchmark average | mIoU: 49.2 -> 49.5 | Improved |
| add STF alone on Pascal VOC 21 (ablation) | MLP baseline · V21 | mIoU: 68.0 -> 66.6 | Hurt |
| increase ODE steps from 5 to 10 (ablation) | DINOde · OVSS 8-benchmark average | mIoU: 48.7 -> 49.5 | Improved |
| increase ODE steps from 10 to 50 (ablation) | DINOde · OVSS 8-benchmark average | mIoU: 49.5 -> 49.6 | No clear effect |
| use ODE alignment vs MLP with smaller DINOv3 ViT-S (ablation) | DINOde · OVSS 8-benchmark average | mIoU: 36.9 -> 38.8 | Improved |
| use ODE alignment vs MLP with DINOv3 ViT-B (ablation) | DINOde · Cityscapes | mIoU: City 38.1 -> 37.7; Avg 44.2 -> 46.1 | Mixed |
| ODE alignment vs MLP alignment on geometry diagnostics (ablation) | STF · class-structure CKA | CKA: 0.843 -> 0.871 | Improved |
| DINOde vs Talk2DINO* with the same DINOv3 backbone | DINOde · OVSS 8-benchmark average | mIoU: 47.6 -> 49.5 | Improved |
Closest papers by meaning
- DINOv2 Meets Text: A Unified Framework for Image- and Pixel-Level Vision-Language Alignment (CVPR 2025 · similarity 0.91)
- Open-Vocabulary Segmentation with Semantic-Assisted Calibration (CVPR 2024 · similarity 0.91)
- Emergent Open-Vocabulary Semantic Segmentation from Off-the-shelf Vision-Language Models (CVPR 2024 · similarity 0.90)
- Leveraging Depth and Language for Open-Vocabulary Domain-Generalized Semantic Segmentation (NeurIPS 2025 · similarity 0.90)
- Distilling Spectral Graph for Object-Context Aware Open-Vocabulary Semantic Segmentation (CVPR 2025 · similarity 0.89)
- Training-free Open-Vocabulary Semantic Segmentation via Diverse Prototype Construction and Sub-region Matching (AAAI 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.
