Xiao Sun, Ymyang, Xinyi Jiang, Yu Tian, Junnan Zhu, Jiang Zhong, Qin Lei, Jingwang Huang, Haoyang Zeng, Xinyu Zhou, Xin Xiao, Kaiwen Wei. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Ymyang, Jiang Zhong, Li Jin, Xiao Sun, Jingwang Huang, Gaojinpeng, Qing Liu, Yang Bai, Jingyuan Zhang, Rui Jiang, Qin Lei, Kaiwen Wei. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Generating sketches that accurately reflect the content of reference images presents numerous challenges. Current methods either require paired training data or fail to accommodate a wider range and diversity of sketch styles. While pre-trained diffusion models have shown strong text-based control capabilities for reference-based content sketch generation, state-of-the-art methods still struggle with reference-based sketch generation for given content. The main difficulties lie in (1) balancing content preservation with style enhancement, and (2) representing content image textures at varying levels of abstraction to approximate the reference sketch style. In this paper, we propose a method (Ref2Sketch-SA) that transforms a given content image into a sketch based on a reference sketch. The core strategies include (1) using DDIM Inversion to enhance structural consistency in the sketch generation of content images; (2) injecting noise into the input image during the denoising process to produce a sketch that retains content attributes while aligning with, yet differing in texture from, the reference. Our model demonstrates superior performance across multiple evaluation metrics, including user style preference.

The zero-crossing-based integrator (ZCBI) offers an energy-efficient alternative to traditional operational transconductance amplifier (OTA)-based switched-capacitor (SC) integrators in analog-to-digital converters (ADCs) such as Delta-Sigma (ΔΣ) and pipeline ADCs, particularly in low supply voltage scenarios. To achieve zero-crossing detection, the ZCBI requires a preset operation to set the output to supply voltage at the start of the integration phase. This preset operation consumes extra dynamic power in each integration phase, and limits the speed and dynamic range of the integrator. This paper introduces a new ZCBI that can determine the direction of zero crossing by a pre-judgement circuit, hence circumventing the need for the preset phase and supporting continuous two-way integration. As a proof of concept, a self-timed incremental zoom ADC using the proposed ZCBI has been fabricated in a standard 180-nm CMOS technology. This ADC consists of a coarse 5-bit SAR ADC and a fine second-order single-bit ΔΣ modulator. The ADC consumes 150 μW from a 1.8-V supply. An SNDR of 77.4 dB is achieved in a 6.25 kHz bandwidth, leading to a Schreier FoM of 154 dB. It is foreseen that implementations in a more advanced technology will further enhance the energy efficiency of the proposed design.

Accurate segmentation of retinal blood vessels is crucial for diagnosing and treating various ophthalmic conditions. Although existing deep learning methods have shown promising results, background pixel noise in images often affects segmentation outcomes to varying degrees. Currently, effectively utilizing features from different domains within an image remains an underexplored yet promising research direction. In this paper, we propose a novel dual-branch network architecture, termed LIONet, which extracts features from RGB images and Local Intensity Order Transform (LIOT) images in two separate branches, and merges them in a Cross-Domain Fusion Module (CDFM). This approach effectively combines the global color information from the RGB domain and the local intensity variation information from the LIOT domain within retinal vessel images. Additionally, we integrate a Residual Excitation Module (REM) into the down-sampling layers of each branch to enhance feature representation and reduce the impact of redundant and irrelevant information. Detailed result analysis and comparisons on three publicly available retinal vessel datasets demonstrate the effectiveness of our approach compared to state-of-the-art methods.

In this paper, we address the significant challenge of data scarcity in the field of crack segmentation, a key aspect of structural health monitoring. To tackle this, we introduce the CrackGrowDiff framework, an innovative approach for expanding crack datasets. Utilizing a two-stage controllable generation process that combines a random walk algorithm and semantic diffusion models, our framework minimizes discrepancy of misalignment between synthetic data and original data while enhancing data informativeness. We further ensure the quality and informativeness of synthetic data through feature space diversity, employing a pre-trained Variational Autoencoder (VAE) for selection based on Kullback-Leibler (KL) divergence. Comparative experiments demonstrate CrackGrowDiff’s superiority over traditional data augmentation and GANs-based methods, making it a substantial advancement in addressing the data scarcity in crack segmentation tasks. A DEMO and related code will be made public: https://huggingface.co/spaces/QinLei086/Two-stage-SDM-for-crack-dataset-expending

Crack segmentation is a critical component in structural health monitoring. Conventional crack segmentation models usually focus on optimizing the cross-entropy-based objective function and overlook the optimization of the subsequent binarization process. In this paper, we redefine the crack segmentation problem as a joint optimization problem, which requires optimizing the binarization process in addition to the objective function of the segmentation model. Simultaneously optimizing both processes demonstrates significant improvements in the model’s segmentation performance. To optimize the binarization process, we propose the Adaptive Dynamic Thresholding Module (ADTM), which reuses the spatial features in the segmentation network to perform an additional regression task to obtain the optimal threshold for each crack image. ADTM is a pluggable component for practical deployment, consuming only a small amount of additional memory during deployment while significantly improving inference accuracy. Experimental results using four different datasets with diverse sources and distributions for crack semantic and instance segmentation demonstrate the effectiveness of ADTM in improving segmentation performance. While ADTM has only been evaluated on cracked data, our findings suggest its potential to improve the performance of other binary classification image segmentation problems.

Biomedical Entity Linking(BEL) is the task of linking biomedical mentions in natural language corpora such as diseases and drugs to standard entities in a given knowledge base. The same biomedical entity can have multiple mentions, including synonyms, morphological variations and names with different word order. Thus, making predictions for each mention with an insufficient context in a single biomedical document is challenging, especially when mentions or linked entities are unseen during training. This paper proposes an inference method for biomedical entity linking based on the heterogeneous graph constructed on multiple documents. We first design a joint representation learning method and compute mention-mention and mention-entity similarity in a unified semantic space. Based on them, we utilize the mutual nearest neighbors relationship between mentions and the relationship between mention and entity to construct a heterogeneous graph. Finally, we apply a clustering-based method to make linking predictions, which associates each mention in the documents with a unique entity. Extensive experiments on three biomedical entity linking benchmarks (MedMentions, BC5CDR and NCBI) demonstrate that our method outperforms other state-of-the-art entity linking models.

Binary medical image segmentation plays a pivotal role in the diagnosis and treatment of a wide range of diseases. However, the performance of the segmentation model is closely related to the choice of the binarization threshold (default 0.5), which is used to binarize the output probability mask. In this paper, we introduce an innovative multi-task learning framework featuring an Adaptive Thresholding Module (ATM) designed to predict the optimal threshold for each image. Within this multi-task learning framework, the segmentation task is divided into two distinct subtasks. The first subtask focuses on the original segmentation task to output the probabilistic mask. Simultaneously, the second subtask leverages spatial features extracted from the segmentation network, with ATM learning and applying these features in a regression task to derive the optimal threshold for each image. Subsequently, a binarization process is enacted using these optimal thresholds, leading to a marked improvement in segmentation accuracy. The crux of ATM’s contribution to enhanced segmentation accuracy lies in its ability to optimize the binarization process, striking a well-balanced equilibrium between the labels of true positive and false positive. We integrated ATM into various medical segmentation models and subjected it to evaluation on datasets encompassing diverse binarized medical image patterns. The results underscore the effectiveness of ATM in elevating the accuracy of pre-existing medical segmentation models.

Document-level relation extraction (DocRE) is an important task in natural language processing, with applications in knowledge graph construction, question answering, and biomedical text analysis. However, existing approaches to DocRE have limitations in predicting relations between entities using fixed entity representations, which can lead to inaccurate results. In this paper, we propose a novel DocRE model that addresses these limitations by using a relation-specific entity representation method and evidence sentence augmentation. Our model uses evidence sentence augmentation to identify top-k evidence sentences for each relation and a relation-specific entity representation method that aggregates the importance of entity mentions using an attention mechanism. These two components work together to capture the context of each entity mention in relation to the specific relation being predicted and select evidence sentences that support accurate relation identification. Finally, we re-predicts entity relations based on the evidence sentences, called relationship reordering module. This module re-predicts entity relationships based on the predicted set of evidence sentences to form k sets of relationship predictions, and then averages these k+1 sets of results to obtain the final relationship predictions. Experimental results on the DocRED dataset demonstrate that our proposed model achieves an F1 score of 62.84% and an lgn F1 score of 60.79%, outperforming state-of-the-art methods.

Current state-of-the-art semantic segmentation methods usually require high computational resources for accurate segmentation. Knowledge distillation has been one promising way to achieve a good trade-off between accuracy and efficiency. However, current distillation methods focus on transferring the spatial relations and ignore the multi-scale context interaction. This paper proposes one novel pixel- region relation distillation (PPRD) to transfer the multi-scale pixel-region relation (PRR) from the teacher to the student. We get the multi-scale regions with pyramid pooling and characterize the multi-scale PRR between the feature and the multi-scale regions. Transferring such PRR from the teacher to the student is beneficial for the student to mimic the teacher better in terms of multi-scale context interaction. Experimental results on two challenging datasets, Cityscapes and Pascal VOC 2012, show that the proposed approach outperforms state-of-the-art distillation methods.

The objective of the concept of T-type Modular Multilevel Converter (TMMC) is to improve the power density of the traditional Modular Multilevel Converter (MMC) exploited in medium voltage motor drive. The size of the major volume contributor “capacitor” can be reduced by 30%~40% with the proposed circuit operating at only half dc link voltage at low frequency low voltage region, which is enabled by an extra full bridge arm inserted between the middle point of the dc and the terminal of the output in each phase leg. It is also found to be a topology that is optimized for usage of SiC device because of its zero requirement for device serialization and its modest dv/dt and di/dt level.