arXiv subjects

Jianwei Xu

Publications and source records attributed to Jianwei Xu.

At least 19 recordsLinked to original sources

arXiv2026

EndoVLM: An Endoscopy Vision-Language Pre-training Model via Anatomy-Guided Sparsity and Progressive Alignment

The development of foundation models (FMs) is crucial for advancing endoscopic image analysis. However, existing endoscopy FMs mainly rely on self-supervised learning from uni-modal images or videos, overlooking the rich semantic knowledge contained in clinical reports. Furthermore, effectively leveraging these records is hindered by a fundamental modality gap: structured anatomical descriptions are not naturally mapped to specific frames within the high-redundancy, uncurated visual streams. In this paper, we present EndoVLM, a novel vision-language FM pre-trained on over 348K endoscopic examinations, each pairing a clinical report with its corresponding image collection. An Anatomy-Guided Sparse Pooling mechanism utilizes textual descriptions as queries to drive sparse attention, efficiently aggregating semantically salient frames into anatomy-specific visual representations across redundant image-sets. Next, a Progressive Semantic-Aware Alignment strategy models clinical taxonomy (anatomy and pathological status) via structured soft targets, bridging the gap from global patient-level matching to fine-grained localized alignment. Finally, a Semantic-Concentrated Masked Autoencoder is applied exclusively to these semantic-rich frames, integrating low-level visual precision with robust high-level semantic representation. Extensive experiments across various downstream tasks demonstrate that EndoVLM outperforms existing foundation models and remains competitive with task-specific methods. Remarkably, EndoVLM also exhibits robust zero-shot generalization capabilities, highlighting its potential for broader clinical application.

cs.CV↗

arXiv2026

E-MRL: Cross-view Aligned Evidence-driven Multimodal Reinforcement Learning for Reliable 3D Tumor Analysis

While Vision-Language Models (VLMs) show great promise in volumetric medical report generation, they frequently suffer from visual hallucinations and a lack of grounding in 3D CT data. Current Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) strategies typically optimize text fidelity alone, essentially rewarding correct diagnoses derived from language priors rather than genuine visual perception. To address this, we propose cross-view aligned Evidence-driven Multimodal Reinforcement Learning (Evidence-MRL, noted as E-MRL), a reliable RL reasoning framework that formulates the generation process as a Markov Decision Process of "diagnosis-localization-verification". Unlike standard approaches, our model is explicitly trained to identify a "key evidence slice" alongside the global diagnostic report, grounding its findings in verifiable visual evidence. Crucially, we introduce a novel cross-view consistency reward, which validates the semantic alignment between the golden-standard report and a local visual re-query of the selected key slice, providing additional rewards for correctly-localized reasoning. Experiments on large-scale 3D CT tumor datasets demonstrate that E-MRL significantly reduces hallucinations and improves diagnostic accuracy compared to SFT and RL baselines, offering a clinically interpretable solution for visually-grounded and tumor analysis.

eess.IV↗

arXiv2025

Imaginarity measures induced by real part states and the complementarity relations

Complex numbers are indispensable in quantum mechanics and the resource theory of imaginarity has been developed recently. In this paper, we propose a method to construct imaginary measures by real part states. Specifically, we propose an imaginarity measure in terms of fidelity and explore its properties. The analytical expression of the imaginarity measure is presented in qubit systems. The relations between the proposed imaginarity measure and some other imaginarity measures (such as geometric imaginarity, Tsallis relative entropy imaginarity and trace norm imaginarity) are derived. The complementarity relations of the imaginarity measure under a complete set of mutually unbiased bases are provided in low-dimensional systems. This work not only highlights the prominent role of the real part state in the imaginarity resource theory, but also reveals the constraint of imaginarity on a complete set of mutually unbiased bases physically.

quant-ph↗

arXiv2025

Bargmann invariants of Gaussian states

Given a set of ordered quantum states, described by density operators $% \{\rho _{j}\}_{j=1}^{n}$, the Bargmann invariant of $\{\rho _{j}\}_{j=1}^{n}$ is defined as tr($\rho _{1}\rho _{2}...\rho _{n}$). Bargmann invariant serves as a fundamental concept for quantum mechanics and has diverse applications in quantum information science. Bosonic Gaussian states are a class of quantum states on infinite-dimensional Hilbert space, widely used in quantum optics and quantum information science. Bosonic Gaussian states are conveniently and conventionally characterized by their means and covariance matrices. In this work, we provide the expression of Bargmann invariant tr($\rho _{1}\rho _{2}...\rho _{n}$) for any $m$-mode bosonic Gaussian states $\{\rho _{j}\}_{j=1}^{n}$ in terms of the means and covariance matrices of $\{\rho _{j}\}_{j=1}^{n}.$ We also use this expression to explore the permissible values of Bargmann invariants for bosonic Gaussian states.

quant-ph↗

arXiv2025

Numerical ranges of Bargmann invariants

Bargmann invariants have recently emerged as powerful tools in quantum information theory, with applications ranging from geometric phase characterization to quantum state distinguishability. Despite their widespread use, a complete characterization of their physically realizable values has remained an outstanding challenge. In this work, we provide a rigorous determination of the numerical range of Bargmann invariants for quantum systems of arbitrary finite dimension. We demonstrate that any permissible value of these invariants can be achieved using either (i) pure states exhibiting circular Gram matrix symmetry or (ii) qubit states alone. These results establish fundamental limits on Bargmann invariants in quantum mechanics and provide a solid mathematical foundation for their diverse applications in quantum information processing.

quant-ph↗

arXiv2024

Hermitian Kirkwood-Dirac real operators for discrete Fourier transformations

The Kirkwood-Dirac (KD) distribution is a quantum state representation that relies on two chosen fixed orthonormal bases, or alternatively, on the transition matrix of these two bases. In recent years, it has been discovered that the KD distribution has numerous applications in quantum information science. The presence of negative or nonreal KD distributions may indicate certain quantum features or advantages. If the KD distribution of a quantum state consists solely of positive or zero elements, the state is called a KD positive state. Consequently, a crucial inquiry arises regarding the determination of whether a quantum state is KD positive when subjected to various physically relevant transition matrices. When the transition matrix is discrete Fourier transform (DFT) matrix of dimension $p$ [\href{https://doi.org/10.1063/5.0164672} {J. Math. Phys. 65, 072201 (2024)}] or $p^{2}$ [\href{https://dx.doi.org/10.1088/1751-8121/ad819a} {J. Phys. A: Math. Theor. 57 435303 (2024)}] with $p$ being prime, it is proved that any KD positive state can be expressed as a convex combination of pure KD positive states. In this work, we prove that when the transition matrix is the DFT matrix of any finite dimension, any KD positive state can be expressed as a real linear combination of pure KD positive states.

quant-ph↗

arXiv2024

Generalizing monocular colonoscopy image depth estimation by uncertainty-based global and local fusion network

Objective: Depth estimation is crucial for endoscopic navigation and manipulation, but obtaining ground-truth depth maps in real clinical scenarios, such as the colon, is challenging. This study aims to develop a robust framework that generalizes well to real colonoscopy images, overcoming challenges like non-Lambertian surface reflection and diverse data distributions. Methods: We propose a framework combining a convolutional neural network (CNN) for capturing local features and a Transformer for capturing global information. An uncertainty-based fusion block was designed to enhance generalization by identifying complementary contributions from the CNN and Transformer branches. The network can be trained with simulated datasets and generalize directly to unseen clinical data without any fine-tuning. Results: Our method is validated on multiple datasets and demonstrates an excellent generalization ability across various datasets and anatomical structures. Furthermore, qualitative analysis in real clinical scenarios confirmed the robustness of the proposed method. Conclusion: The integration of local and global features through the CNN-Transformer architecture, along with the uncertainty-based fusion block, improves depth estimation performance and generalization in both simulated and real-world endoscopic environments. Significance: This study offers a novel approach to estimate depth maps for endoscopy images despite the complex conditions in clinic, serving as a foundation for endoscopic automatic navigation and other clinical tasks, such as polyp detection and segmentation.

cs.CV↗

arXiv2024

Coherence and imaginarity of quantum states

Baumgratz, Cramer and Plenio established a rigorous framework (BCP framework) for quantifying the coherence of quantum states [\href{http://dx.doi.org/10.1103/PhysRevLett.113.140401}{Phys. Rev. Lett. 113, 140401 (2014)}]. In BCP framework, a quantum state is called incoherent if it is diagonal in the fixed orthonormal basis, and a coherence measure should satisfy some conditions. For a fixed orthonormal basis, if a quantum state $ρ$ has nonzero imaginary part, then $ρ$ must be coherent. How to quantitatively characterize this fact? In this work, we show that any coherence measure $C$ in BCP framework has the property $C(ρ)-C($Re$ρ)\geq 0$ if $C$ is invariant under state complex conjugation, i.e., $C(ρ)=C(ρ^{\ast })$, here $ρ^{\ast }$ is the conjugate of $ρ,$ Re$ρ$ is the real part of $ρ.$ If $C$ does not satisfy $C(ρ)=C(ρ^{\ast }),$ we can define a new coherence measure $C^{\prime }(ρ)=\frac{1}{2}[C(ρ)+C(ρ^{\ast })]$ such that $C^{\prime }(ρ)=C^{\prime }(ρ^{\ast }).$ We also establish some similar results for bosonic Gaussian states.

quant-ph↗

arXiv2024

Kirkwood-Dirac classical pure states

Kirkwood-Dirac (KD) distribution is a representation of quantum states. Recently, KD distribution has been employed in many scenarios such as quantum metrology, quantum chaos and foundations of quantum theory. KD distribution is a quasiprobability distribution, and negative or nonreal elements may signify quantum advantages in certain tasks. A quantum state is called KD classical if its KD distribution is a probability distribution. Since most quantum information processings use pure states as ideal resources, then a key problem is to determine whether a quantum pure state is KD classical. In this paper, we provide some characterizations for the general structure of KD classical pure states. As an application of our results, we prove a conjecture raised by De Bièvre [Phys. Rev. Lett. 127, 190404 (2021)] which finds out all KD classical pure states for discrete Fourier transformation.

quant-ph↗

arXiv2023

Quantifying the imaginarity of quantum states via Tsallis relative entropy

It is a fundamental question that why quantum mechanics uses complex numbers instead of only real numbers. To address this topic, recently, a rigorous resource theory for the imaginarity of quantum states were established, and several imaginarity measures were proposed. In this work, we propose a new imaginarity measure based on the Tsallis relative entropy. This imaginarity measure has explicit expression, and also, it is computable for bosonic Gaussian states.

quant-ph↗

arXiv2023

Imaginarity of Gaussian states

It has been a long-standing debate that why quantum mechanics uses complex numbers but not only real numbers. To address this topic, in recent years, the imaginarity theory has been developed in the way of quantum resource theory. However, the existing imaginarity theory mainly focuses on the quantum systems with finite dimensions. Gaussian states are widely used in many fields of quantum physics, but they are in the quantum systems with infinite dimensions. In this paper we establish a resource theory of imaginarity for bosonic Gaussian states. To do so, under the Fock basis, we determine the real Gaussian states and real Gaussian channels in terms of the means and covariance matrices of Gaussian states. Also, we provide two imaginary measures for Gaussian states based on the fidelity.

quant-ph↗

arXiv2023

Two-dimensional layered materials meet perovskite oxides: A combination for high-performance electronic devices

As the Si-based transistors scale down to atomic dimensions, the basic principle of current electronics, which heavily relies on the tunable charge degree of freedom, faces increasing challenges to meet the future requirements of speed, switching energy, heat dissipation, packing density as well as functionalities. Heterogeneous integration, where dissimilar layers of materials and functionalities are unrestrictedly stacked at an atomic scale, is appealing to next-generation electronics, such as multi-functional, neuromorphic, spintronic and ultra-low power devices, because it unlocks technologically useful interfaces of distinct functionalities. Recently, the combination of functional perovskite oxides and the two-dimensional layered materials (2DLMs) led to unexpected functionalities and enhanced device performance. In this review, we review the recent progress of the heterogeneous integration of perovskite oxides and 2DLMs from the perspectives of fabrication and interfacial properties, electronic applications, challenges as well as outlooks. In particular, we focus on three types of attractive applications, namely field-effect transistors, memory, and neuromorphic electronics. The van der Waals integration approach is extendible to other oxides and 2DLMs, leading to almost unlimited combinations of oxides and 2DLMs and contributing to future high-performance electronic and spintronic devices.

cond-mat.mtrl-sci↗

arXiv2023

Quantifying the phase of quantum states

Phase is a basic ingredient for quantum states since quantum mechanics uses complex numbers to describe quantum states. In this letter, we introduce a rigorous framework to quantify the phase of quantum states. To do so, we regard phase as a quantum resource, and specify the free states and free operations. We determine the conditions a phase measure should satisfy and provide some phase measures. We also propose the notion of intrinsic phase for quantum states.

quant-ph↗

arXiv2022

Classification of incompatibility for two orthonormal bases

For two orthonormal bases of a $d$-dimensional complex Hilbert space, the notion of complete incompatibility was introduced recently by De Bièvre [Phys. Rev. Lett. 127, 190404 (2021)]. In this work, we introduce the notion of $s$-order incompatibility with positive integer $s$ satisfying $2\leq s\leq d+1.$ In particular, $(d+1)$-order incompatibility just coincides with the complete incompatibility. We establish some relations between $s$-order incompatibility, minimal support uncertainty and rank deficiency of the transition matrix. As an example, we determine the incompatibility order of the discrete Fourier transform with any finite dimension.

quant-ph↗

arXiv2022

Real-time automatic polyp detection in colonoscopy using feature enhancement module and spatiotemporal similarity correlation unit

Automatic detection of polyps is challenging because different polyps vary greatly, while the changes between polyps and their analogues are small. The state-of-the-art methods are based on convolutional neural networks (CNNs). However, they may fail due to lack of training data, resulting in high rates of missed detection and false positives (FPs). In order to solve these problems, our method combines the two-dimensional (2-D) CNN-based real-time object detector network with spatiotemporal information. Firstly, we use a 2-D detector network to detect static images and frames, and based on the detector network, we propose two feature enhancement modules-the FP Relearning Module (FPRM) to make the detector network learning more about the features of FPs for higher precision, and the Image Style Transfer Module (ISTM) to enhance the features of polyps for sensitivity improvement. In video detection, we integrate spatiotemporal information, which uses Structural Similarity (SSIM) to measure the similarity between video frames. Finally, we propose the Inter-frame Similarity Correlation Unit (ISCU) to combine the results obtained by the detector network and frame similarity to make the final decision. We verify our method on both private databases and publicly available databases. Experimental results show that these modules and units provide a performance improvement compared with the baseline method. Comparison with the state-of-the-art methods shows that the proposed method outperforms the existing ones which can meet real-time constraints. It's demonstrated that our method provides a performance improvement in sensitivity, precision and specificity, and has great potential to be applied in clinical colonoscopy.

cs.CV↗

arXiv2021

$l_{1}$ norm of coherence is not equal to its convex roof quantifier

Since a rigorous framework for quantifying quantum coherence was established by Baumgratz et al. [T. Baumgratz, M. Cramer, and M. B. Plenio, Phys. Rev. Lett. 113, 140401 (2014)], many coherence measures have been found. For a given coherence measure $C$, extending the values of $C$ on pure states to mixed states by the convex roof construction, we will get a valid coherence measure $\overline{C}$, we call $\overline{C}$ the corresponding convex roof quantifier of $C$. Whether $C=\overline{C}$ for a given coherence measure is an important question. In this work, we show that for the widely used coherence measure, $l_{1}$ norm of coherence $C_{l_{1}}$, it holds that $C_{l_{1}}\neq \overline{C_{l_{1}}}$.

quant-ph↗

arXiv2021

Estimating coherence with respect to general quantum measurements

The conventional coherence is defined with respect to a fixed orthonormal basis, i.e., to a von Neumann measurement. Recently, generalized quantum coherence with respect to general positive operator-valued measurements (POVMs) has been presented. Several well-defined coherence measures, such as the relative entropy of coherence $C_{r}$, the $l_{1}$ norm of coherence $C_{l_{1}}$ and the coherence $C_{T,α}$ based on Tsallis relative entropy with respect to general POVMs have been obtained. In this work, we investigate the properties of $C_{r}$, $l_{1}$ and $C_{T,α}$. We estimate the upper bounds of $C_{l_{1}}$; we show that the minimal error probability of the least square measurement state discrimination is given by $C_{T,1/2}$; we derive the uncertainty relations given by $C_{r}$, and calculate the average values of $C_{r}$, $C_{T,α}$ and $C_{l_{1}}$ over random pure quantum states. All these results include the corresponding results of the conventional coherence as special cases.

quant-ph↗

arXiv2021

Self-Adaptive Transfer Learning for Multicenter Glaucoma Classification in Fundus Retina Images

The early diagnosis and screening of glaucoma are important for patients to receive treatment in time and maintain eyesight. Nowadays, deep learning (DL) based models have been successfully used for computer-aided diagnosis (CAD) of glaucoma from retina fundus images. However, a DL model pre-trained using a dataset from one hospital center may have poor performance on a dataset from another new hospital center and therefore its applications in the real scene are limited. In this paper, we propose a self-adaptive transfer learning (SATL) strategy to fill the domain gap between multicenter datasets. Specifically, the encoder of a DL model that is pre-trained on the source domain is used to initialize the encoder of a reconstruction model. Then, the reconstruction model is trained using only unlabeled image data from the target domain, which makes the encoder in the model adapt itself to extract useful high-level features both for target domain images encoding and glaucoma classification, simultaneously. Experimental results demonstrate that the proposed SATL strategy is effective in the domain adaptation task between one private and two public glaucoma diagnosis datasets, i.e. pri-RFG, REFUGE, and LAG. Moreover, the proposed strategy is completely independent of the source domain data, which meets the real scene application and the privacy protection policy.

eess.IV↗