Research map / Image generation / Text-to-image diffusion
WAS: Dataset and Methods for Artistic Text Segmentation
Xudong Xie, Yuzhe Li, Yang Liu, Zhifei Zhang, Zhaowen Wang, Wei Xiong, 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 layer-wise momentum query to Mask2Former decoder (ablation) | Mask2Former · WAS-R test | fgIoU: 80.21 -> 80.95 | Improved |
| add skeleton-assisted head (ablation) | Mask2Former + LMQ · WAS-R test | fgIoU: 80.95 -> 82.11 | Improved |
| pre-train on synthetic WAS-S then fine-tune on WAS-R (ablation) | WASNet · WAS-R test | fgIoU: 82.11 -> 84.18 | Improved |
| pre-train on WAS-S | Mask2Former · WAS-R test | fgIoU: 80.21 -> 82.42 | Improved |
| increase synthetic data from 5w to 10w images (ablation) | WASNet pre-training · WAS-R test | fgIoU: 83.25 -> 84.18 | Improved |
| increase synthetic data to 20w images (ablation) | WASNet pre-training · WAS-R test | fgIoU: 84.18 -> 84.01 | No clear effect |
| use BLIP2 instead of Monkey for captioning (ablation) | WAS-S generation · WAS-R test | fgIoU: 84.18 -> 83.07 | Hurt |
| use 1000 fonts instead of 250 artistic fonts (ablation) | WAS-S mask render · WAS-R test | fgIoU: 84.18 -> 82.35 | Hurt |
| momentum coefficient alpha from 0.8 to 0.1 (ablation) | WASNet LMQ · WAS-R test | fgIoU: 82.11 -> 81.31 | Hurt |
| momentum coefficient 0.9 vs 0.8 (ablation) | WASNet LMQ · WAS-R test | fgIoU: 82.11 -> 82.03 | No clear effect |
| WASNet vs TextFormer (uses extra box labels) on Total-Text | WASNet (pre-train) · Total-Text | fgIoU: 82.10 vs 79.30 | Hurt |
Closest papers by meaning
- Advancing Manga Analysis: Comprehensive Segmentation Annotations for the Manga109 Dataset (CVPR 2025 · similarity 0.86)
- Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing (NeurIPS 2024 · similarity 0.85)
- Text Image Inpainting via Global Structure-Guided Diffusion Models (AAAI 2024 · similarity 0.85)
- Layout-Agnostic Scene Text Image Synthesis with Diffusion Models (CVPR 2024 · similarity 0.85)
- DreamText: High Fidelity Scene Text Synthesis (CVPR 2025 · similarity 0.85)
- Alchemist: Turning Public Text-to-Image Data into Generative Gold (NeurIPS 2025 · similarity 0.84)
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.
