Research map / Vision-language and multimodal / Human and embodied modeling
Embedding Rotation Invariance for Provable Multi-Oriented Scene Text Recognition
zhibin ma, Pengwen Dai, Yi Liu, Xugong Qin, Chenyun Yu, Xiaochun Cao
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 rotation-equivariant encoder (F-Conv in RELG) (ablation) | RISTER-S · IC13 avg over 4 rotations | accuracy: 96.98 -> 98.40 | Improved |
| add rotation-invariant decoder (RITD) instead of CTC (ablation) | RISTER-S · IC13 avg over 4 rotations | accuracy: 93.82 -> 98.40 | Improved |
| add random-rotation augmentation to the non-equivariant-encoder variant (ablation) | w/o Rot-E Encoder · IC13 avg over 4 rotations | accuracy: 96.98 -> 97.25 | Improved |
| add random-rotation augmentation to the CTC-decoder variant (ablation) | w/o Rot-I Decoder · IC13 avg over 4 rotations | accuracy: 93.82 -> 95.48 | Improved |
| add random-rotation augmentation (ablation) | w/o Rot-E Encoder · IC13 0 degrees | accuracy: 98.4 -> 97.5 | Hurt |
| add F-Conv equivariant convolutions to ResNet45 (ablation) | ResNet45 encoder with RITD · Union14M-B Multi-Oriented | accuracy: 90.8 -> 94.9 | Improved |
| add G-CNN equivariant layers to ResNet45 (ablation) | ResNet45 encoder with RITD · Union14M-B Multi-Oriented | accuracy: 90.8 -> 90.9 | No clear effect |
| add Stand-Alone equivariant self-attention to ViT (ablation) | ViT encoder with RITD · IC13 0 degrees | accuracy: 98.5 -> 98.3 | Mixed |
| add self-attention global blocks to RELG (ablation) | RELG · SVT 0 degrees | accuracy: 96.4 -> 98.3 | Improved |
| add self-attention global blocks to RELG (ablation) | RELG · Union14M-B Multi-Oriented | accuracy: 95.0 -> 96.6 | Improved |
| increase input resolution from 32x128 to 128x128 (ablation) | SVTRv2 · Mul-Ori | accuracy: 89.5 -> 90.2 | Improved |
| increase input resolution from 32x128 to 128x128 (ablation) | IGTR · Mul-Ori | accuracy: 92.6 -> 92.9 | Improved |
Closest papers by meaning
- Rotation-Sensitive Regression for Oriented Scene Text Detection (CVPR 2018 · similarity 0.87)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.87)
- RobustScanner: Dynamically Enhancing Positional Clues for Robust Text Recognition (ECCV 2020 · similarity 0.86)
- Revisiting Scene Text Recognition: A Data Perspective (ICCV 2023 · similarity 0.86)
- AON: Towards Arbitrarily-Oriented Text Recognition (CVPR 2018 · similarity 0.86)
- Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing (NeurIPS 2024 · 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.
