Self-supervised learning has emerged as a powerful approach for leveraging large-scale unlabeled data to improve model performance in various domains. In this paper, we explore masked self-supervised pre-training for text recognition transformers. Specifically, we propose two modifications to the pre-training phase: progressively increasing the masking probability, and modifying the loss function to incorporate both masked and non-masked patches. We conduct extensive experiments using a dataset of 50M unlabeled text lines for pre-training and four differently sized annotated datasets for fine-tuning. Furthermore, we compare our pre-trained models against those trained with transfer learning, demonstrating the effectiveness of the self-supervised pre-training. In particular, pre-training consistently improves the character error rate of models, in some cases up to 30 % relatively. It is also on par with transfer learning but without relying on extra annotated text lines.

Scaling architectures have been proven effective for improving Scene Text Recognition (STR), but the individual contribution of vision encoder and text decoder scaling remain under-explored. In this work, we present an in-depth empirical analysis and demonstrate that, contrary to previous observations, scaling the decoder yields significant performance gains, always exceeding those achieved by encoder scaling alone. We also identify label noise as a key challenge in STR, particularly in real-world data, which can limit the effectiveness of STR models. To address this, we propose Cloze Self-Distillation (CSD), a method that mitigates label noise by distilling a student model from context-aware soft predictions and pseudolabels generated by a teacher model. Additionally, we enhance the decoder architecture by introducing differential cross-attention for STR. Our methodology achieves state-of-the-art performance on 10 out of 11 benchmarks using only real data, while significantly reducing the parameter size and computational costs.

Handwritten Paragraph Text Recognition (HPTR) is a challenging task in Computer Vision, requiring the transformation of a paragraph text image, rich in handwritten text, into text encoding sequences. One of the most advanced models for this task is Vertical Attention Network (VAN), which utilizes a Vertical Attention Module (VAM) to implicitly segment paragraph text images into text lines, thereby reducing the difficulty of the recognition task. However, from a network structure perspective, VAM is a single-branch module, which is less effective in learning compared to multi-branch modules. In this paper, we propose a new module, named Re-parameterizing Vertical Attention Fusion Module (RVAFM), which incorporates structural re-parameterization techniques. RVAFM decouples the structure of the module during training and inference stages. During training, it uses a multi-branch structure for more effective learning, and during inference, it uses a single-branch structure for faster processing. The features learned by the multi-branch structure are fused into the single-branch structure through a special fusion method named Re-parameterization Fusion (RF) without any loss of information. As a result, we achieve a Character Error Rate (CER) of 4.44% and a Word Error Rate (WER) of 14.37% on the IAM paragraph-level test set. Additionally, the inference speed is slightly faster than VAN.

Mainstream Scene Text Recognition (STR) algorithms are developed based on RGB cameras which are sensitive to challenging factors such as low illumination, motion blur, and cluttered backgrounds. In this paper, we propose to recognize the scene text using bio-inspired event cameras by collecting and annotating a large-scale benchmark dataset, termed EventSTR. It contains 9,928 high-definition (1280 * 720) event samples and involves both Chinese and English characters. We also benchmark multiple STR algorithms as the baselines for future works to compare. In addition, we propose a new event-based scene text recognition framework, termed SimC-ESTR. It first extracts the event features using a visual encoder and projects them into tokens using a Q-former module. More importantly, we propose to augment the vision tokens based on a memory mechanism before feeding into the large language models. A similarity-based error correction mechanism is embedded within the large language model to correct potential minor errors fundamentally based on contextual information. Extensive experiments on the newly proposed EventSTR dataset and two simulation STR datasets fully demonstrate the effectiveness of our proposed model. We believe that the dataset and algorithmic model can innovatively propose an event-based STR task and are expected to accelerate the application of event cameras in various industries. The source code and pre-trained models will be released on https://github.com/Event-AHU/EventSTR

Method and Novelty: The authors present a model that has a number of strengths. First, the character-level model is trained on synthetically generated images from a font library, independently of the training corpus. Second, the model converts each training image into a factor graph and learns the spatial relationships between landmarks in each character. This model can readily assign a probability to each candidate character for an image, and the authors provide a description of a two-stage inference algorithm that consists of approximate belief propagation followed by refinement via a backtracking procedure. The candidate characters are then supplied to a word model, which is a fairly standard structured prediction using bigram and trigram features.

Recent advances in Handwritten Text Recognition (HTR) have led to significant reductions in transcription errors on standard benchmarks under the i.i.d. assumption, thus focusing on minimizing in-distribution (ID) errors. However, this assumption does not hold in real-world applications, which has motivated HTR research to explore Transfer Learning and Domain Adaptation techniques. In this work, we investigate the unaddressed limitations of HTR models in generalizing to out-of-distribution (OOD) data. We adopt the challenging setting of Domain Generalization, where models are expected to generalize to OOD data without any prior access. To this end, we analyze 336 OOD cases from eight state-of-the-art HTR models across seven widely used datasets, spanning five languages. Additionally, we study how HTR models leverage synthetic data to generalize. We reveal that the most significant factor for generalization lies in the textual divergence between domains, followed by visual divergence. We demonstrate that the error of HTR models in OOD scenarios can be reliably estimated, with discrepancies falling below 10 points in 70\% of cases. We identify the underlying limitations of HTR models, laying the foundation for future research to address this challenge.

The Multi-language Video Subtitle Dataset is a comprehensive collection designed to support research in text recognition across multiple languages. This dataset includes 4,224 subtitle images extracted from 24 videos sourced from online platforms. It features a wide variety of characters, including Thai consonants, vowels, tone marks, punctuation marks, numerals, Roman characters, and Arabic numerals. With 157 unique characters, the dataset provides a resource for addressing challenges in text recognition within complex backgrounds. It addresses the growing need for high-quality, multilingual text recognition data, particularly as videos with embedded subtitles become increasingly dominant on platforms like YouTube and Facebook. The variability in text length, font, and placement within these images adds complexity, offering a valuable resource for developing and evaluating deep learning models. The dataset facilitates accurate text transcription from video content while providing a foundation for improving computational efficiency in text recognition systems. As a result, it holds significant potential to drive advancements in research and innovation across various computer science disciplines, including artificial intelligence, deep learning, computer vision, and pattern recognition.

The segmentation-free research efforts for addressing handwritten text recognition can be divided into three categories: connectionist temporal classification (CTC), hidden Markov model and encoder-decoder methods. In this paper, inspired by the above three modeling methods, we propose a new recognition network by using a novel three-dimensional (3D) attention module and global-local context information. Based on the feature maps of the last convolutional layer, a series of 3D blocks with different resolutions are split. Then, these 3D blocks are fed into the 3D attention module to generate sequential visual features. Finally, by integrating the visual features and the corresponding global-local context features, a well-designed representation can be obtained. Main canonical neural units including attention mechanisms, fully-connected layer, recurrent unit and convolutional layer are efficiently organized into a network and can be jointly trained by the CTC loss and the cross-entropy loss. Experiments on the latest Chinese handwritten text datasets (the SCUT-HCCDoc and the SCUT-EPT) and one English handwritten text dataset (the IAM) show that the proposed method can make a new milestone.

We introduce "Generative Fusion Decoding" (GFD), a novel shallow fusion framework, utilized to integrate Large Language Models (LLMs) into multi-modal text recognition systems such as automatic speech recognition (ASR) and optical character recognition (OCR). We derive the formulas necessary to enable GFD to operate across mismatched token spaces of different models by mapping text token space to byte token space, enabling seamless fusion during the decoding process. The framework is plug-and-play, compatible with various auto-regressive models, and does not require re-training for feature alignment, thus overcoming limitations of previous fusion techniques. We highlight three main advantages of GFD: First, by simplifying the complexity of aligning different model sample spaces, GFD allows LLMs to correct errors in tandem with the recognition model, reducing computation latencies. Second, the in-context learning ability of LLMs is fully capitalized by GFD, increasing robustness in long-form speech recognition and instruction aware speech recognition. Third, GFD enables fusing recognition models deficient in Chinese text recognition with LLMs extensively trained on Chinese. Our evaluation demonstrates that GFD significantly improves performance in ASR and OCR tasks, with ASR reaching state-of-the-art in the NTUML2021 benchmark. GFD provides a significant step forward in model integration, offering a unified solution that could be widely applicable to leveraging existing pre-trained models through step by step fusion.

Internal Language Model (LM)-based methods use permutation language modeling (PLM) to solve the error correction caused by conditional independence in external LM-based methods. However, random permutations of human interference cause fit oscillations in the model training, and Iterative Refinement (IR) operation to improve multimodal information decoupling also introduces additional overhead. To address these issues, this paper proposes the Hierarchical Attention autoregressive Model with Adaptive Permutation (HAAP) to enhance the location-context-image interaction capability, improving autoregressive generalization with internal LM. First, we propose Implicit Permutation Neurons (IPN) to generate adaptive attention masks to dynamically exploit token dependencies. The adaptive masks increase the diversity of training data and prevent model dependency on a specific order. It reduces the training overhead of PLM while avoiding training fit oscillations. Second, we develop Cross-modal Hierarchical Attention mechanism (CHA) to couple context and image features. This processing establishes rich positional semantic dependencies between context and image while avoiding IR. Extensive experimental results show the proposed HAAP achieves state-of-the-art (SOTA) performance in terms of accuracy, complexity, and latency on several datasets.