PulseAugur
EN
LIVE 00:54:06

New 2D positional encoding boosts Transformer scene text recognition

By PulseAugur Editorial · [2 sources] ·

Researchers have developed a novel 2D Rotary Position Embedding (2D-RoPE-STR) method to improve Transformer-based scene text recognition (STR). This new approach addresses the limitations of existing 1D positional encodings by better capturing the 2D spatial structure of text images, which is crucial for handling curved, rotated, and perspective-distorted text. The method introduces anisotropic dimension allocation and extends rotary coupling into encoder-decoder cross-attention, enabling more accurate autoregressive decoding. Evaluations on six standard benchmarks show significant gains, particularly on irregular text layouts. AI

IMPACT Enhances Transformer models' ability to recognize text in complex, real-world image conditions.

RANK_REASON The cluster contains an academic paper detailing a new method for scene text recognition.

Read on arXiv cs.CV →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

New 2D positional encoding boosts Transformer scene text recognition

COVERAGE [2]

  1. arXiv cs.CV TIER_1 English(EN) · Zobeir Raisi ·

    2D Rotary Position Embedding for Scene Text Recognition with Transformers

    arXiv:2607.13458v1 Announce Type: new Abstract: Scene Text Recognition (STR) remains challenging due to the diversity of text appearances, including curvature, rotation, and perspective distortion. Recent Transformer-based approaches perform well but usually rely on one-dimension…

  2. arXiv cs.CV TIER_1 English(EN) · Zobeir Raisi ·

    2D Rotary Position Embedding for Scene Text Recognition with Transformers

    Scene Text Recognition (STR) remains challenging due to the diversity of text appearances, including curvature, rotation, and perspective distortion. Recent Transformer-based approaches perform well but usually rely on one-dimensional positional encodings that ignore the 2D spati…

RELATED ENTITIES

RELATED TOPICS