Research map / Video / Video-language understanding
VimTS: A Unified Video and Image Text Spotter for Enhancing the Cross-Domain Generalization
Yuliang Liu, Mingxin Huang, Hao Yan, Linger Deng, Weijia Wu, Hao Lü, Chunhua Shen, Lianwen Jin, Xiang Bai
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 PQGM to joint word+line full-tuning (ablation) | VimTS tuning · TotalText-word -> IC15 | end-to-end H-mean (%): 62.5 -> 67.6 | Improved |
| add PQGM to joint word+line full-tuning (ablation) | VimTS tuning · CTW1500-line -> TotalText-line | end-to-end H-mean (%): 36.6 -> 54.0 | Improved |
| add PQGM to joint word+line full-tuning (ablation) | VimTS tuning · TotalText-word -> CTW1500-word | end-to-end H-mean (%): 71.1 -> 81.5 | Improved |
| naively merge word- and line-level data without PQGM (ablation) | single-task baseline · CTW1500-line -> TotalText-line | end-to-end H-mean (%): 50.5 -> 36.6 | Hurt |
| replace full-tuning with task-aware adapter (3.0% params) (ablation) | VimTS + PQGM · TotalText-word -> IC15 | end-to-end H-mean (%): 67.6 -> 68.0 | Improved |
| replace standard adapter with task-aware adapter (ablation) | VimTS + PQGM · TotalText-word -> IC15 | end-to-end H-mean (%): 67.2 -> 68.0 | Improved |
| aggregate detection information before recognition in adapter (ablation) | task-aware adapter · TotalText-word -> IC15 | end-to-end H-mean (%): 67.5 -> 68.0 | Improved |
| add synthetic VTD-368k to image-level training (ablation) | VimTS · ICDAR2015 video | MOTA (%): 61.0 -> 61.8 | Improved |
| add real ICDAR2015 video data on top of VTD-368k (ablation) | VimTS · ICDAR2015 video | MOTA (%): 61.8 -> 62.3 | Improved |
| add VTD-368k under stricter IoU 0.7 (ablation) | VimTS · DSText v2 | MOTA (%): -14.4 -> -1.3 | Improved |
| add VTD-368k to TransDETR (ablation) | TransDETR · ICDAR2015 video | ID_F1 (%): 72.8 -> 74.0 | Improved |
| add VTD-368k to TransVTSpotter (ablation) | TransVTSpotter · ICDAR2015 video | ID_F1 (%): 61.1 -> 62.5 | Improved |
Closest papers by meaning
- GoMatching: A Simple Baseline for Video Text Spotting via Long and Short Term Matching (NeurIPS 2024 · similarity 0.89)
- SemiETS: Integrating Spatial and Content Consistencies for Semi-Supervised End-to-end Text Spotting (CVPR 2025 · similarity 0.86)
- TOPA: Extending Large Language Models for Video Understanding via Text-Only Pre-Alignment (NeurIPS 2024 · similarity 0.86)
- ViTEraser: Harnessing the Power of Vision Transformers for Scene Text Removal with SegMIM Pretraining (AAAI 2024 · similarity 0.86)
- Domain Generalization in CLIP via Learning with Diverse Text Prompts (CVPR 2025 · similarity 0.85)
- Mimir: Improving Video Diffusion Models for Precise Text Understanding (CVPR 2025 · similarity 0.85)
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.
