Research map / Detection and segmentation / Open-set and few-shot detection
In Defense of Lazy Visual Grounding for Open-Vocabulary Semantic Segmentation
Dahyun Kang, Minsu Cho
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 |
|---|---|---|---|
| discover object masks first with Panoptic cut then ground text, instead of pixel-wise grounding | SCLIP · VOC21 | mIoU: 59.1 -> 62.1 | Improved |
| same two-stage approach on ADE20K | SCLIP · ADE20K (150 classes) | mIoU: 16.1 -> 15.8 | Hurt |
| iterate Panoptic cut rather than a single Normalized cut (ablation) | LaVG · VOC21 | mIoU: 47.6 -> 62.1 | Improved |
| use Panoptic cut instead of CutLER masks (ablation) | LaVG · VOC21 | mIoU: 57.3 -> 62.1 | Improved |
| raise max Panoptic cut iterations from 4 to 16 (ablation) | LaVG · VOC21 | mIoU: 61.1 -> 62.1 | Improved |
| raise max Panoptic cut iterations from 16 to 32 (ablation) | LaVG · VOC21 | mIoU: 62.1 -> 61.4 | No clear effect |
| average patch embeddings in the object mask (prototype) instead of O-masked attention (ablation) | LaVG · VOC20 | mIoU: 61.7 -> 82.5 | Improved |
| average patch embeddings in the object mask instead of Gaussian blur on non-object (ablation) | LaVG · COCO-Object | mIoU: 31.5 -> 34.2 | Improved |
| use DINO instead of SCLIP backbone for Panoptic cut (ablation) | LaVG · COCO-Object | mIoU: 23.2 -> 34.2 | Improved |
| add sliding-window inference (ablation) | LaVG · VOC21 | mIoU: 32.3 -> 62.1 | Improved |
| run iterative graph partitioning before grounding | LaVG · inference time | inference time (s): 12.59 vs 0.54 (CATseg) | Hurt |
Closest papers by meaning
- Your Large Vision-Language Model Only Needs A Few Attention Heads For Visual Grounding (CVPR 2025 · similarity 0.90)
- Grounding Everything: Emerging Localization Properties in Vision-Language Transformers (CVPR 2024 · similarity 0.90)
- GroundVLP: Harnessing Zero-Shot Visual Grounding from Vision-Language Pre-training and Open-Vocabulary Object Detection (AAAI 2024 · similarity 0.89)
- Emergent Open-Vocabulary Semantic Segmentation from Off-the-shelf Vision-Language Models (CVPR 2024 · similarity 0.89)
- Mask Grounding for Referring Image Segmentation (CVPR 2024 · similarity 0.89)
- Segment Anyword: Mask Prompt Inversion for Open-Set Grounded Segmentation (ICML 2025 · similarity 0.88)
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.
