Research map / Detection and segmentation / Open-set and few-shot detection
Learning To Generate Text-Grounded Mask for Open-World Semantic Segmentation From Only Image-Text Pairs
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Learn region-text alignment directly with text-grounded contrastive learning instead of image-text alignment | TCL · 8 semantic segmentation datasets | mIoU: +7.8 average mIoU | Helped |
| Add a grounding decoder without TCL loss (ablation) | MaskCLIP-style baseline · PASCAL VOC20 training split | mIoU: 53.2 -> 52.3 | Hurt |
| Replace conventional CL with full TCL losses (ablation) | CL baseline · PASCAL VOC20 training split | mIoU: 61.1 -> 77.4 | Helped |
| Use image-level TCL alone with area loss (ablation) | TCL · PASCAL VOC20 training split | mIoU: 61.1 -> 74.6 | Helped |
| Use feature-level TCL alone with area loss (ablation) | TCL · PASCAL VOC20 training split | mIoU: 61.1 -> 76.0 | Helped |
| Use both image-level and feature-level TCL losses with area loss (ablation) | TCL · PASCAL VOC20 training split | mIoU: 76.0 -> 77.4 | Helped |
| Add conventional CL alongside TCL losses (ablation) | TCL · PASCAL VOC20 training split | mIoU: 77.4 -> 75.6 | Hurt |
| Remove area TCL loss (ablation) | TCL · PASCAL VOC20 training split | mIoU: 77.4 -> 67.1 | Hurt |
| Set TCL loss weight to 1.0 rather than the default 0.1 (ablation) | TCL · PASCAL VOC20 training split | mIoU: 77.4 -> 68.2 | Hurt |
| Remove TV smooth regularization (ablation) | TCL · PASCAL VOC20 training split | mIoU: 77.4 -> 73.8 | Hurt |
| Apply PAMR refinement (ablation) | TCL · 8-dataset average | mIoU: +3.3 mIoU | Helped |
| Apply PAMR refinement (ablation) | MaskCLIP · 8-dataset average | mIoU: -1.5 mIoU | Hurt |
Closest papers by meaning
- Image-Text Co-Decomposition for Text-Supervised Semantic Segmentation (CVPR 2024 · similarity 0.90)
- Training-Free Open-Vocabulary Segmentation with Offline Diffusion-Augmented Prototype Generation (CVPR 2024 · similarity 0.88)
- Segment Anyword: Mask Prompt Inversion for Open-Set Grounded Segmentation (ICML 2025 · similarity 0.88)
- Text4Seg: Reimagining Image Segmentation as Text Generation (ICLR 2025 · similarity 0.87)
- Emergent Open-Vocabulary Semantic Segmentation from Off-the-shelf Vision-Language Models (CVPR 2024 · similarity 0.87)
- CLIP as RNN: Segment Countless Visual Concepts without Training Endeavor (CVPR 2024 · similarity 0.87)
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.