AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

Rethinking Text Segmentation: A Novel Dataset and a Text-Specific Refinement Approach

Xingqian Xu, Zhifei Zhang, Zhaowen Wang, Brian Price, Zhonghao Wang, Humphrey Shi

CVPR 2021 · Detection and segmentation › Open-set and few-shot detection · approach: Interactive and referring segmentation · 78 citations (OpenAlex)

Check 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 changedWhereMeasuredEffect
Apply TexRNet refinement to HRNetV2-W48HRNetV2-W48 · TextSegfgIoU: 85.98 -> 86.84Improved
Apply TexRNet refinement to HRNetV2-W48HRNetV2-W48 · ICDAR13 FSTfgIoU: 72.45 -> 73.38Improved
Apply TexRNet refinement to DeeplabV3+DeeplabV3+ · COCO_TSfgIoU: 72.07 -> 73.98Improved
Apply TexRNet refinement to DeeplabV3+DeeplabV3+ · MLT_SfgIoU: 84.63 -> 86.31Improved
Apply TexRNet refinement to HRNetV2-W48HRNetV2-W48 · Total-TextfgIoU: 76.23 -> 78.47Improved
Add key features pooling and attention (ablation)TexRNet (base) with DeeplabV3+ · TextSegfgIoU: 84.86 -> 85.36Improved
Add trimap loss (ablation)TexRNet with attention · TextSegfgIoU: 85.36 -> 85.55Improved
Add glyph discriminator (ablation)TexRNet with attention and trimap loss · TextSegfgIoU: 85.55 -> 86.06Improved
Train with TextSeg in addition to ICDAR13 FST, COCO_TS, and MLT_STexRNet · ICDAR13 FST testfgIoU: 76.68 -> 78.65Improved
Train with TextSeg in addition to Total-Text, COCO_TS, and MLT_STexRNet · Total-Text testfgIoU: 77.40 -> 80.01Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy