Recent advances demonstrate that reinforcement learning with verifiable rewards (RLVR) significantly enhances the reasoning capabilities of large language models (LLMs). However, standard RLVR faces challenges with reward sparsity, where zero rewards from consistently incorrect candidate answers provide no learning signal, particularly in challenging tasks. To address this,we propose Multi-Expert Mutual Learning GRPO (MEML-GRPO), an innovative framework that utilizes diverse expert prompts as system prompts to generate a broader range of responses, substantially increasing the likelihood of identifying correct solutions. Additionally, we introduce an inter-expert mutual learning mechanism that facilitates knowledge sharing and transfer among experts, further boosting the model’s performance through RLVR. Extensive experiments across multiple reasoning benchmarks show that MEML-GRPO delivers significant improvements, achieving an average performance gain of 4.89% with Qwen and 11.33% with Llama, effectively overcoming the core limitations of traditional RLVR methods.

The advent of Large Multimodal Models (LMMs) has fueled extensive research due to their sophisticated reasoning capabilities. However, for understanding text-rich images, challenges persist in fully leveraging the potential of LMMs, and existing methods struggle with effectively processing high-resolution images. Addressing this, we introduce TextCoT, a training-free Chain-of-Thought framework that improves text-rich image understanding by leveraging LMMs’ captioning abilities for global context and detailed local textual analysis. TextCoT comprises three stages: Global Context Generation, Macro-Scale Positioning, and Fine-Grained Visual Inspection—each contributing to a comprehensive understanding and precise information extraction needed for accurate question-answering. Our method requires no additional training, offering immediate plug-and-play functionality. We have demonstrated TextCoT’s effectiveness and adaptability across various benchmarks. The source code is available at https://github.com/bzluan/TextCoT .

An-Lan Wang, Jingqun Tang, Lei Liao, Hao Feng, Qi Liu, Xiang Fei, Jinghui Lu, Han Wang, Hao Liu, Yuliang Liu, Xiang Bai, Can Huang. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Scoring the Optical Character Recognition (OCR) capabilities of Large Multimodal Models (LMMs) has witnessed growing interest. Existing benchmarks have highlighted the impressive performance of LMMs in text recognition; however, their abilities in certain challenging tasks, such as text localization, handwritten content extraction, and logical reasoning, remain underexplored. To bridge this gap, we introduce OCRBench v2, a large-scale bilingual text-centric benchmark with currently the most comprehensive set of tasks ($4\times$ more tasks than the previous multi-scene benchmark OCRBench), the widest coverage of scenarios ($31$ diverse scenarios), and thorough evaluation metrics, with $10,000$ human-verified question-answering pairs and a high proportion of difficult samples. Moreover, we construct a private test set with $1,500$ manually annotated images. The consistent evaluation trends observed across both public and private test sets validate the OCRBench v2's reliability. After carefully benchmarking state-of-the-art LMMs, we find that most LMMs score below $50$ ($100$ in total) and suffer from five-type limitations, including less frequently encountered text recognition, fine-grained perception, layout perception, complex element parsing, and logical reasoning. The benchmark and evaluation scripts are available at https://github.com/Yuliang-Liu/MultimodalOCR.

In video lane detection, there are rich temporal contexts among successive frames, which is under-explored in existing lane detectors. In this work, we propose LaneTCA to bridge the individual video frames and explore how to effectively aggregate the temporal context. Technically, we develop an accumulative attention module and an adjacent attention module to abstract the long-term and short-term temporal context, respectively. The accumulative attention module continuously accumulates visual information during the journey of a vehicle, while the adjacent attention module propagates this lane information from the previous frame to the current frame. The two modules are meticulously designed based on the transformer architecture. Finally, these long-short context features are fused with the current frame features to predict the lane lines in the current frame. Extensive quantitative and qualitative experiments are conducted on two prevalent benchmark datasets. The results demonstrate the effectiveness of our method, achieving several new state-of-the-art records. The codes and models are available athttps://github.com/Alex-1337/LaneTCA.

In this work, we present DeepEraser, an effective deep network for generic text removal. DeepEraser utilizes a recurrent architecture that erases the text in an image via iterative operations. Our idea comes from the process of erasing pencil script, where the text area designated for removal is subject to continuous monitoring and the text is attenuated progressively, ensuring a thorough and clean erasure. Technically, at each iteration, an innovative erasing module is deployed, which not only explicitly aggregates the previous erasing progress but also mines additional semantic context to erase the target text. Through iterative refinements, the text regions are progressively replaced with more appropriate content and finally converge to a relatively accurate status. Furthermore, a custom mask generation strategy is introduced to improve the capability of DeepEraser for adaptive text removal, as opposed to indiscriminately removing all the text in an image. Our DeepEraser is notably compact with only 1.4 M parameters and trained in an end-to-end manner. To verify its effectiveness, extensive experiments are conducted on several prevalent benchmarks, including SCUT-Syn, SCUT-EnsText, and Oxford Synthetic text dataset. The quantitative and qualitative results demonstrate the effectiveness of our DeepEraser over the state-of-the-art methods, as well as its strong generalization ability in custom mask text removal.

Pre-trained vision-language models (VLMs) have shown remarkable generalization capabilities via prompting, which leverages VLMs as knowledge bases to extract information beneficial for downstream tasks. However, existing methods primarily employ uni-modal prompting, which only engages a uni-modal branch, failing to simultaneously adjust vision-language (V-L) features. Additionally, the one-pass forward pipeline in VLM encoding struggles to align V-L features that have a huge gap. Confronting these challenges, we propose a novel method, Progressive Multi-modal conditional Prompt Tuning (ProMPT). ProMPT exploits a recurrent structure, optimizing and aligning V-L features by iteratively utilizing image and current encoding information. It comprises an initialization and a multi-modal iterative evolution (MIE) module. Initialization is responsible for encoding images and text using a VLM, followed by a feature filter that selects text features similar to image. MIE then facilitates multi-modal prompting through class-conditional vision prompting, instance-conditional text prompting, and feature filtering. In each MIE iteration, vision prompts are obtained from filtered text features via a vision generator, promoting image features to focus more on target object during vision prompting. The encoded image features are fed into a text generator to produce text prompts that are more robust to class shifts. Thus, V-L features are progressively aligned, enabling advance from coarse to exact prediction. Extensive experiments are conducted in three settings to evaluate the efficacy of ProMPT. The results indicate that ProMPT outperforms existing methods on average across all settings, demonstrating its superior generalization and robustness. Code is available at https://github.com/qiuxiaoyu9954/ProMPT.

Tables contain factual and quantitative data accompanied by various structures and contents that pose challenges for machine comprehension. Previous methods generally design task-specific architectures and objectives for individual tasks, resulting in modal isolation and intricate workflows. In this paper, we present a novel large vision-language model, TabPedia, equipped with a concept synergy mechanism. In this mechanism, all the involved diverse visual table understanding (VTU) tasks and multi-source visual embeddings are abstracted as concepts. This unified framework allows TabPedia to seamlessly integrate VTU tasks, such as table detection, table structure recognition, table querying, and table question answering, by leveraging the capabilities of large language models (LLMs). Moreover, the concept synergy mechanism enables table perception-related and comprehension-related tasks to work in harmony, as they can effectively leverage the needed clues from the corresponding source perception embeddings. Furthermore, to better evaluate the VTU task in real-world scenarios, we establish a new and comprehensive table VQA benchmark, ComTQA, featuring approximately 9,000 QA pairs. Extensive quantitative and qualitative experiments on both table perception and comprehension tasks, conducted across various public benchmarks, validate the effectiveness of our TabPedia. The superior performance further confirms the feasibility of using LLMs for understanding visual tables when all concepts work in synergy. The benchmark ComTQA has been open-sourced at https://huggingface.co/datasets/ByteDance/ComTQA. The source code and model also have been released at https://github.com/zhaowc-ustc/TabPedia.

Contour-based instance segmentation has been actively studied, thanks to its flexibility and elegance in processing visual objects within complex backgrounds. In this work, we propose a novel deep network architecture,i.e., PolySnake, for generic contour-based instance segmentation. Motivated by the classic Snake algorithm, the proposed PolySnake achieves superior and robust segmentation performance with an iterative and progressive contour refinement strategy. Technically, PolySnake introduces a recurrent update operator to estimate the object contour iteratively. It maintains a single estimate of the contour that is progressively deformed toward the object boundary. At each iteration, PolySnake builds a semantic-rich representation for the current contour and feeds it to the recurrent operator for further contour adjustment. Through the iterative refinements, the contour progressively converges to a stable status that tightly encloses the object instance. Beyond the scope of general instance segmentation, extensive experiments are conducted to validate the effectiveness and generalizability of our PolySnake in two additional specific task scenarios, including scene text detection and lane detection. The results demonstrate that the proposed PolySnake outperforms the existing advanced methods on several multiple prevalent benchmarks across the three tasks. The codes and pre-trained models are available at https://github.com/fh2019ustc/PolySnake.

In recent years, the proliferation of smartphones has led to an upsurge in the digitization of document files via these portable devices. However, images captured by smartphones often suffer from distortions, thereby negatively affecting digital preservation and downstream applications. To address this issue, we introduce DRNet, a novel deep network for document image rectification. Our approach is based on three key designs. Firstly, we exploit the intrinsic geometric consistency inherent in document images to guide the learning process of distortion rectification. Secondly, we design a coarse-to-fine rectification network to leverage the representations derived from the distorted document image, thereby enhancing the rectification result. Thirdly, we propose a unique perspective for supervising the learning of rectification networks, where undistorted document images are employed for supervision, which is free of warping mesh as ground truth in existing methods. Technically, both low-level pixel alignment and high-level semantic alignment jointly contribute to the learning of the mapping relationship between deformed document images and distortion-free ones. We evaluate our method on the challenging DocUNet Benchmark dataset, where it sets a series of state-of-the-art records, demonstrating the superiority of our approach compared to existing learning-based solutions. Additionally, we conduct a comprehensive series of ablation experiments to further validate the effectiveness and merits of our method.

In recent years, tremendous efforts have been made on document image rectification, but existing advanced algorithms are limited to processing restricted document images, i.e., the input images must incorporate a complete document. Once the captured image merely involves a local text region, its rectification quality is degraded and unsatisfactory. Our previously proposed DocTr, a transformer-assisted network for document image rectification, also suffers from this limitation. In this work, we present DocTr++, a novel unified framework for document image rectification, without any restrictions on the input distorted images. Our major technical improvements can be concluded in three aspects. Firstly, we upgrade the original architecture by adopting a hierarchical encoder-decoder structure for multi-scale representation extraction and parsing. Secondly, we reformulate the pixel-wise mapping relationship between the unrestricted distorted document images and the distortion-free counterparts. The obtained data is used to train our DocTr++ for unrestricted document image rectification. Thirdly, we contribute a real-world test set and metrics applicable for evaluating the rectification quality. To our best knowledge, this is the first learning-based method for the rectification of unrestricted document images. Extensive experiments are conducted, and the results demonstrate the effectiveness and superiority of our method. We hope our DocTr++ will serve as a strong baseline for generic document image rectification, prompting the further advancement and application of learning-based algorithms.

In fisheye images, rich distinct distortion patterns are regularly distributed in the image plane. These distortion patterns are independent of the visual content and provide informative cues for rectification. To make the best of such rectification cues, we introduce SimFIR, a simple framework for fisheye image rectification based on self-supervised representation learning. Technically, we first split a fisheye image into multiple patches and extract their representations with a Vision Transformer (ViT). To learn fine-grained distortion representations, we then associate different image patches with their specific distortion patterns based on the fisheye model, and further subtly design an innovative unified distortion-aware pretext task for their learning. The transfer performance on the downstream rectification task is remarkably boosted, which verifies the effectiveness of the learned representations. Extensive experiments are conducted, and the quantitative and qualitative results demonstrate the superiority of our method over the state-of-the-art algorithms as well as its strong generalization ability on real-world fisheye images.

This paper presents IP-SLT, a simple yet effective framework for sign language translation (SLT). Our IP-SLT adopts a recurrent structure and enhances the semantic representation (prototype) of the input sign language video via an iterative refinement manner. Our idea mimics the behavior of human reading, where a sentence can be digested repeatedly, till reaching accurate understanding. Technically, IP-SLT consists of feature extraction, prototype initialization, and iterative prototype refinement. The initialization module generates the initial prototype based on the visual feature extracted by the feature extraction module. Then, the iterative refinement module leverages the cross-attention mechanism to polish the previous prototype by aggregating it with the original video feature. Through repeated refinement, the prototype finally converges to a more stable and accurate state, leading to a fluent and appropriate translation. In addition, to leverage the sequential dependence of prototypes, we further propose an iterative distillation loss to compress the knowledge of the final iteration into previous ones. As the autoregressive decoding process is executed only once in inference, our IP-SLT is ready to improve various SLT systems with acceptable overhead. Extensive experiments are conducted on public benchmarks to demonstrate the effectiveness of the IP-SLT.

Tremendous efforts have been made on document image rectification, but how to learn effective representation of such distorted images is still under-explored. In this paper, we present DocMAE, a novel self-supervised framework for document image rectification. Our motivation is to encode the structural cues in document images by leveraging masked autoencoder to benefit the rectification, i.e., the document boundaries, and text lines. Specifically, we first mask random patches of the background-excluded document images and then reconstruct the missing pixels. With such a self-supervised learning approach, the network is encouraged to learn the intrinsic structure of deformed documents by restoring document boundaries and missing text lines. Transfer performance in the downstream rectification task validates the effectiveness of our method. Extensive experiments are conducted to demonstrate the effectiveness of our method.

Camera lenses often suffer from optical aberrations, causing radial distortion in the captured images. In those images, there exists a clear and general physical distortion model. However, in existing solutions, such rich geometric prior is under-utilized, and the formulation of an effective prediction target is under-explored. To this end, we introduce Radial Distortion TRansformer (RDTR), a new framework for radial distortion rectification. Our RDTR includes a model-aware pre-training stage for distortion feature extraction and a deformation estimation stage for distortion rectification. Technically, on the one hand, we formulate the general radial distortion (i.e., barrel distortion and pincushion distortion) in camera-captured images with a shared geometric distortion model and perform a unified model-aware pre-training for its learning. With the pre-training, the network is capable of encoding the specific distortion pattern of a radially distorted image. After that, we transfer the learned representations to the learning of distortion rectification. On the other hand, we introduce a new prediction target called backward warping flow for rectifying images with any resolution while avoiding image defects. Extensive experiments are conducted on our synthetic dataset, and the results demonstrate that our method achieves state-of-the-art performance while operating in real-time. Besides, we also validate the generalization of RDTR on real-world images. Our source code and the proposed dataset are publicly available at https://github.com/wwd-ustc/RDTR.

In this work, we propose a new framework, called Document Image Transformer (DocTr), to address the issue of geometry and illumination distortion of the document images. Specifically, DocTr consists of a geometric unwarping transformer and an illumination correction transformer. By setting a set of learned query embedding, the geometric unwarping transformer captures the global context of the document image by self-attention mechanism and decodes the pixel-wise displacement solution to correct the geometric distortion. After geometric unwarping, our illumination correction transformer further removes the shading artifacts to improve the visual quality and OCR accuracy. Extensive evaluations are conducted on several datasets, and superior results are reported against the state-of-the-art methods. Remarkably, our DocTr achieves $20.02%$ Character Error Rate (CER), a $15%$ absolute improvement over the state-of-the-art methods. Moreover, it also shows high efficiency on running time and parameter count.