Research map / Detection and segmentation / Open-set and few-shot detection
Global Knowledge Calibration for Fast Open-Vocabulary Segmentation
Kunyang Han, Yong Liu, Jun Hao Liew, Henghui Ding, Jiajun Liu, Yitong Wang, Yansong Tang, Yujiu Yang, Jiashi Feng, Yao Zhao, Yunchao Wei
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 |
|---|---|---|---|
| add text diversification (synonyms) to training (ablation) | Global Knowledge Calibration (CLIP R-50) · Cityscapes | mIoU: 27.61 -> 32.16 | Improved |
| add text-guided knowledge distillation (ablation) | Global Knowledge Calibration (CLIP R-50) · Cityscapes | mIoU: 27.61 -> 32.62 | Improved |
| combine TD and TGKD (ablation) | Global Knowledge Calibration (CLIP R-50) · Pascal Context | mIoU: 39.70 -> 41.91 | Improved |
| use text-guided instead of vision-guided distillation (ablation) | Global Knowledge Calibration · Pascal Context | mIoU: 39.67 -> 41.23 | Improved |
| use vision-guided distillation over all regions (vs no KD) (ablation) | Global Knowledge Calibration · Pascal Context | mIoU: 39.70 -> 39.67 | No clear effect |
| add text diversification to Simbaseline (ablation) | Simbaseline · Cityscapes | mIoU: 30.0 -> 34.7 | Improved |
| vary synonym sampling strategy (Random, GroupAvg, GroupMax) (ablation) | Text diversification · Pascal Context | mIoU: 41.45 / 41.80 / 41.17 | No clear effect |
| use spatial-token teacher and pre-projection student embeddings (ablation) | Text-guided KD · Cityscapes | mIoU: 30.43 -> 32.62 | Improved |
| add text-guided KD to video baseline (ablation) | Video open-vocab baseline · VIPSeg unseen | unseen mIoU: 2.4 -> 2.9 | Improved |
| add text diversification on top of KD for video (ablation) | Video open-vocab baseline · VIPSeg unseen | unseen mIoU: 2.9 -> 8.5 | Improved |
| use ImageNet-pretrained backbone instead of CLIP (ablation) | Global Knowledge Calibration · Pascal Context | mIoU: 41.9 -> 45.6 | No clear effect |
| drop the extra frozen CLIP encoder at inference | Global Knowledge Calibration · Computational cost (R101) | FLOPs: 1165.07G -> 151.44G | Improved |
Closest papers by meaning
- Open-Vocabulary Segmentation with Semantic-Assisted Calibration (CVPR 2024 · similarity 0.93)
- Emergent Open-Vocabulary Semantic Segmentation from Off-the-shelf Vision-Language Models (CVPR 2024 · similarity 0.92)
- Transferable and Principled Efficiency for Open-Vocabulary Segmentation (CVPR 2024 · similarity 0.92)
- Unveiling the Knowledge of CLIP for Training-Free Open-Vocabulary Semantic Segmentation (AAAI 2025 · similarity 0.91)
- Dual Semantic Guidance for Open Vocabulary Semantic Segmentation (CVPR 2025 · similarity 0.91)
- Training-free Open-Vocabulary Semantic Segmentation via Diverse Prototype Construction and Sub-region Matching (AAAI 2025 · similarity 0.90)
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.
