Diffusion Model Research Articles

Mathematical Models of Diffusion in Physiology.

Diffusion is a mass transport phenomenon caused by chaotic thermal movements of molecules. Studying the transport in specific domain is simplified by using evolutionary differential equations for local concentration of the molecules instead of complete information on molecular paths [1]. Compounds in a fluid mixture tend to smooth out its spatial concentration inhomogeneities by diffusion. Rate of the transport is proportional to the concentration gradient and coefficient of diffusion of the compound in ordinary diffusion. The evolving concentration profile c(x,t) is then solution of evolutionary partial differential equation deltac/deltat=DDeltac where D is diffusion coefficient and Delta is Laplacian operator. Domain of the equation may be a region in space, plane or line, a manifold, such as surface embedded in space, or a graph. The Laplacian operates on smooth functions defined on given domain. We can use models of diffusion for such diverse tasks as: a) design of method for precise measurement of receptors mobility in plasmatic membrane by confocal microscopy [2], b) evaluation of complex geometry of trabeculae in developing heart [3] to show that the conduction pathway within the embryonic ventricle is determined by geometry of the trabeculae.

UIEDP: Boosting underwater image enhancement with diffusion prior

Underwater image enhancement (UIE) aims to generate clear images from low-quality underwater images. Due to the unavailability of clear reference images, researchers often synthesize them to construct paired datasets for training deep models. However, these synthesized images may sometimes lack quality, adversely affecting training outcomes. To address this issue, we propose to boost UIE with Diffusion Prior (UIEDP). It is a novel framework treating UIE as a posterior distribution sampling process of clear images conditioned on degraded underwater inputs. Specifically, UIEDP combines a pre-trained diffusion model capturing natural image priors with any existing UIE algorithm, leveraging the latter to guide conditional generation. The diffusion prior mitigates the drawbacks of inferior synthetic images, resulting in higher-quality image generation. Extensive experiments have demonstrated that our UIEDP yields significant improvements across various metrics, especially no-reference image quality assessment. And the generated enhanced images also exhibit a more natural appearance.

Temporally Coherent Video Cartoonization for Animation Scenery Generation

The automatic transformation of short background videos from real scenarios into other forms with a visually pleasing style, like those used in cartoons, holds application in various domains. These include animated films, video games, advertisements, and many other areas that involve visual content creation. A method or tool that can perform this task would inspire, facilitate, and streamline the work of artists and people who produce this type of content. This work proposes a method that integrates multiple components to translate short background videos into other forms that contain a particular style. We apply a fine-tuned latent diffusion model with an image-to-image setting, conditioned with the image edges (computed with holistically nested edge detection) and CLIP-generated prompts to translate the keyframes from a source video, ensuring content preservation. To maintain temporal coherence, the keyframes are translated into grids and the style is interpolated with an example-based style propagation algorithm. We quantitatively assess the content preservation and temporal coherence using CLIP-based metrics over a new dataset of 20 videos translated into three distinct styles.

Discrepancy-based diffusion models for lesion detection in brain MRI

Diffusion probabilistic models (DPMs) have exhibited significant effectiveness in computer vision tasks, particularly in image generation. However, their notable performance heavily relies on labelled datasets, which limits their application in medical images due to the associated high-cost annotations. Current DPM-related methods for lesion detection in medical imaging, which can be categorized into two distinct approaches, primarily rely on image-level annotations. The first approach, based on anomaly detection, involves learning reference healthy brain representations and identifying anomalies based on the difference in inference results. In contrast, the second approach, resembling a segmentation task, employs only the original brain multi-modalities as prior information for generating pixel-level annotations. In this paper, our proposed model – discrepancy distribution medical diffusion (DDMD) – for lesion detection in brain MRI introduces a novel framework by incorporating distinctive discrepancy features, deviating from the conventional direct reliance on image-level annotations or the original brain modalities. In our method, the inconsistency in image-level annotations is translated into distribution discrepancies among heterogeneous samples while preserving information within homogeneous samples. This property retains pixel-wise uncertainty and facilitates an implicit ensemble of segmentation, ultimately enhancing the overall detection performance. Thorough experiments conducted on the BRATS2020 benchmark dataset containing multimodal MRI scans for brain tumour detection demonstrate the great performance of our approach in comparison to state-of-the-art methods.

SingVisio: Visual analytics of diffusion model for singing voice conversion

In this study, we present SingVisio, an interactive visual analysis system that aims to explain the diffusion model used in singing voice conversion. SingVisio provides a visual display of the generation process in diffusion models, showcasing the step-by-step denoising of the noisy spectrum and its transformation into a clean spectrum that captures the desired singer’s timbre. The system also facilitates side-by-side comparisons of different conditions, such as source content, melody, and target timbre, highlighting the impact of these conditions on the diffusion generation process and resulting conversions. Through comparative and comprehensive evaluations, SingVisio demonstrates its effectiveness in terms of system design, functionality, explainability, and user-friendliness. It offers users of various backgrounds valuable learning experiences and insights into the diffusion model for singing voice conversion.

Fractionation by Spatially Heterogeneous Diffusion: Experiments and the Two-Component Random Walk Model.

The fundamental question regarding the fractionation phenomenon is whether diffusion alone is responsible for it or whether an additional advection dynamic is involved. We studied the fractionation by diffusion of particles in spatially heterogeneous environments. By experimentally observing the time-sequential fractionation patterns of dye particles diffusing across a solid-solid interface of varying polyacrylamide gel densities, we found that the two-component diffusion model accurately captures the observed fractionation dynamics. In contrast, single-component diffusion models by Fick, Wereide, and Chapman do not. Our results indicate that diffusion alone can explain the fractionation phenomenon and that additional advection dynamics are not involved. The underlying physics in the fractionation phenomenon is discussed by using a two-component random walk model.

Chain forming diatoms use different strategies to avoid diffusion limited N assimilation

AbstractFormation of large colonies by phytoplankton is considered a disadvantage during low nutrient and non‐turbulent conditions because of diffusion limitation and competition by neighboring cells. This is assumed by diffusion models and not empirical measurements. Here, we measured cell‐specific dissolved inorganic carbon (DIC) and nitrate () assimilation in two chain‐forming Skeletonema marinoi strains by combining secondary ion mass spectrometry with stable isotopic tracer incubations. The pelagic strains were recently germinated from resting stages and thus not adapted to high nutrient regimes of laboratory cultures. During the exponential phase, the cells assimilated excess relative to DIC. The DIC and assimilation varied with chain length or the position of the cell within chains but did not follow any consistent trend. Solitary cells were rare and did not exceed 6% of all cells. During the nitrogen‐limited stationary phase, assimilation was lower than modeled by mass transfer theory at diffusion limitation. assimilation rates were apparently limited by the biological uptake rate at nitrate concentrations < 0.46 μmol L−1. We conclude that S. marinoi is adapted to high nutrient concentrations and may seldom be diffusion limited in coastal waters with ambient concentrations > 0.46 μmol L−1. These findings contrast those previously reported in larger chain forming diatoms, for example, Chaetoceros, which appear to circumvent diffusion limitation at low ambient concentrations and succeed S. marinoi during spring blooms.

Text-guided image-to-sketch diffusion models

Recently, with the continuous advancement of deep learning techniques, research on sketch synthesis has been progressing. However, existing methods still face challenges in generating human-like freehand sketches from real-world natural images at both object and scene levels. To address this, we propose SketchDiffusion, a text-guided freehand sketch synthesis method based on conditional stable diffusion. In SketchDiffusion, we design a novel image enhancing module to efficiently extract high-quality image features. Moreover, we utilize additional guidance from global and local features extracted by a U-shaped diffusion guidance network to control the noise addition and denoising process of the diffusion model, thereby significantly improving controllability and performance in freehand sketch synthesis. Beyond the model architecture, we leverage the designed BLIP-based text generation method to create 70,280 text prompts for foreground, background, and panorama sketch synthesis in the extensive SketchyCOCO dataset, thereby improving the overall effectiveness of model training. Compared to the state-of-the-art methods, our proposed SketchDiffusion has shown an average improvement of over 16.4%, 16.75%, and 12.8% on three quantitative metrics (sketch recognition, sketch-based retrieval, and user perceptual study), respectively. Furthermore, our approach not only excels in synthesizing freehand sketches containing multiple abstract objects but also has multiple applications in supporting human–computer interaction.

Empowering Natural Medicine: Nanocarrier-based Oral Delivery of Vigna Radiata Extract for Effective Diabetes Management

Background: Poor solubility and low oral bioavailability are major challenges associated with the oral delivery of the antidiabetic drug Vigna radiata (VR). Nanostructured lipid carriers (NLCs) have emerged as a promising strategy to overcome these limitations and improve the therapeutic efficacy of VR. This study investigated the potential of NLCs for VR delivery and explored the influence of formulation parameters on NLC properties and drug release behavior. Method: NLCs loaded with VR were prepared using the melt emulsion ultrafiltration technique. The effect of two key formulation variables – the ratio of liquid lipid to solid lipid and the concentration of the surfactant – were investigated in terms of particle size, zeta potential, and drug encapsulation efficiency. The in vitro release profiles of the VR-NLC formulations were evaluated, and the optimal formulation was subjected to further analysis to investigate its release kinetics. Result: The NLCs exhibited particle sizes ranging from 108.9 to 192.3 nm and all formulations possessed a negative zeta potential (-3.68 to -10.9 mV), indicating good stability and potential for resisting aggregation. Interestingly, the lowest solid lipid to liquid lipid ratio and the lowest surfactant concentration yielded the highest drug encapsulation efficiency, highlighting the complex interplay between these factors. All VR-NLC formulations exhibited a biphasic, time-dependent in vitro release pattern, suggesting an initial burst release followed by a sustained release phase. This biphasic profile is promising for achieving both rapid onset of action and long-lasting glycemic control, which are crucial aspects of effective diabetes management. The optimized NLC formulation showed an in vitro release pattern that adhered to the Higuchi diffusion model, suggesting a controlled release mechanism where the drug diffuses steadily out of the NLC matrix. This finding indicates potentially predictable and consistent drug delivery in vivo. Conclusion: This study demonstrates the potential of NLCs as a promising platform for the controlled oral delivery of VR. NLCs can overcome the inherent limitations of VR and provide a convenient and effective oral antidiabetic option for patients. Further research is needed to confirm the efficacy and safety of NLC-encapsulated VR in vivo using relevant animal models. This will pave the way for the development of a novel and potentially transformative treatment option for diabetes.

Numerical simulation of deuterium retention in tungsten under ELM-like conditions

Retention of hydrogen isotopes (HI) in plasma-facing components (PFCs) is a crucial process influencing the operation of a fusion device. The dynamics of HI retention is mainly determined by irradiation conditions of PFCs. These conditions can change due to the onset of fast transient events, such as edge-localised modes (ELMs). The development of ELMs results in repetitive short-term plasma bursts on PFCs, affecting the exposure regime. In this work, the effect of ELM-like loads on the fuel retention is numerically analysed using a one-dimensional diffusion model, implemented in the FESTIM code. As a representative simulation case, the deuterium (D) retention in tungsten (W) is considered with the geometry and inter-ELM exposure conditions relevant for large fusion devices. Temporal dependencies of heat and particle fluxes during intra-ELM stages are accounted for by applying the free-streaming model for an ELM filament transport. The simulations were conducted for three inter-ELM exposure regimes with various properties of transient loads, D trapping sites, and D recombination rates. Compared to the ELM-free irradiation, the onset of transients is shown to mainly reduce the D retention rate because of significant material heating induced by the arrival of energetic ELMy particles. This relative difference can decrease if the material is characterised by strong trapping sites or limited desorption from a surface. The findings also demonstrate that additional heating during transients provides conditions for a faster D migration into the bulk. In addition, we discuss the possibility of using ELM-average loads to obtain quick assessments of the D content. The approach is shown to be applicable for the case of small ELMs and is used to derive the analytical expressions for the D distribution in W within the steady-state approximation.

Achieving the United Nations sustainable development goals – innovation diffusion and business model innovations

Purpose The study aims to analyse the race towards green development and United Nations sustainable development goals (SDGs) in the cases of Huawei and Shell. Both companies are the leaders in their respective industries. Huawei is an example case study representing the information and communications technology (ICT) industry whereas Shell is an example case study representing the oil and gas industry. The research analyses of the races in achieving UN SDGs were undertaken based on the innovation diffusion framework with the use of machine learning algorithms trained to extract data on sustainability activities and initiatives. Design/methodology/approach The research analyses the two case studies of Huawei and Shell. The research was undertaken through the steps of training machine learning algorithms, industry benchmarking and evaluating the performance of the race. The analyses regarding the activities and initiatives of Huawei and Shell in contributing towards SDGs are based on the data in the past 10 years (Years 2010–2019) using machine learning to extract data on sustainability activities and initiatives. In the case of Huawei, 313 sustainability reports were fed to the unsupervised machine learning algorithms revealing 15,101 sustainability actions and initiatives related to UN SDGs in the ICT industry. In the case of Shell, 2,015 sustainability reports were fed to the unsupervised machine learning algorithms revealing 47,365 sustainability actions and initiatives related to UN SDGs in the oil and gas industry. Findings The analyses of findings revealed that Huawei and Shell performed very well in progressing towards the UN SDGs. Huawei had strong performance in the ICT industry with regard to SDGs No. 3, 4, 7, 8, 11, 12 and 16 while Shell had strong performance in the oil and gas industry with regard to SDGs No. 3, 4, 6, 7, 8, 12 and 16. Both companies had placed a focus on achieving SDG 12 responsible consumption and production, SDG 7 affordable and clean energy and SDG 4 quality education. The synthesised business model innovations of Huawei and Shell had shown their environmental, social and governance strategies – Huawei’s 2030 vision for green development and Shell’s 2050 vision for net zero emissions. Practical implications The five pillars of people, planet, prosperity, peace and partnership according to the UN 2030 agenda for sustainable development have shown the way a company operates to promote sustainable eco-systems. The extent to which both Huawei and Shell link corporate strategies to the UN SDGs has reflected their implementation progress. Furthermore, the business model innovations of Huawei and Shell provides a useful framework which can be applied to encourage other companies/organisations in various industries to undertake ESG activities in practice. Originality/value The main contribution of this research is the application of machine learning algorithms and the innovation diffusion model in analysing the SDGs performance. The study applies the innovation diffusion framework to explore strategic actions and initiatives of Huawei and Shell in transitioning towards sustainability. The use of machine learning algorithms has identified their sustainability approach in achieving the UN SDGs.

Enhancing Underwater Object Detection and Classification Using Advanced Imaging Techniques: A Novel Approach with Diffusion Models

Underwater object detection and classification pose significant challenges due to environmental factors such as water turbidity and variable lighting conditions. This research proposes a novel approach that integrates advanced imaging techniques with diffusion models to address these challenges effectively, aligning with Sustainable Development Goal (SDG) 14: Life Below Water. The methodology leverages the Convolutional Block Attention Module (CBAM), Modified Swin Transformer Block (MSTB), and Diffusion model to enhance the quality of underwater images, thereby improving the accuracy of object detection and classification tasks. This study utilizes the TrashCan dataset, comprising diverse underwater scenes and objects, to validate the proposed method’s efficacy. This study proposes an advanced imaging technique YOLO (you only look once) network (AIT-YOLOv7) for detecting objects in underwater images. This network uses a modified U-Net, which focuses on informative features using a convolutional block channel and spatial attentions for color correction and a modified swin transformer block for resolution enhancement. A novel diffusion model proposed using modified U-Net with ResNet understands the intricate structures in images with underwater objects, which enhances detection capabilities under challenging visual conditions. Thus, AIT-YOLOv7 net precisely detects and classifies different classes of objects present in this dataset. These improvements are crucial for applications in marine ecology research, underwater archeology, and environmental monitoring, where precise identification of marine debris, biological organisms, and submerged artifacts is essential. The proposed framework advances underwater imaging technology and supports the sustainable management of marine resources and conservation efforts. The experimental results demonstrate that state-of-the-art object detection methods, namely SSD, YOLOv3, YOLOv4, and YOLOTrashCan, achieve mean accuracies (mAP@0.5) of 57.19%, 58.12%, 59.78%, and 65.01%, respectively, whereas the proposed AIT-YOLOv7 net reaches a mean accuracy (mAP@0.5) of 81.4% on the TrashCan dataset, showing a 16.39% improvement. Due to this improvement in the accuracy and efficiency of underwater object detection, this research contributes to broader marine science and technology efforts, promoting the better understanding and management of aquatic ecosystems and helping to prevent and reduce the marine pollution, as emphasized in SDG 14.

BADM: Boundary-Assisted Diffusion Model for Skin Lesion Segmentation

Skin lesion segmentation is one of the basic tasks in the field of medical image analysis. Deep learning approaches have emerged as crucial tools for segmenting skin lesion images, and their accuracy in processing abnormalities influences the ultimate diagnostic outcomes. Skin lesion images suffer from problems such as variability in shape and size and fuzzy boundaries between lesions and background, making it difficult to accurately segment lesions from images. To address this challenge, we propose Boundary-assisted Diffusion Model for Skin Lesion Segmentation that incorporates boundary information into the denoising learning process of the diffusion model by introducing an additional boundary-aware task. We design a Boundary-Aware Fusion Model to capture boundary information from deep networks and incorporate boundary information into the learning process of the diffusion model. A Scale-Coordinated Interaction Model is proposed, which effectively transmits and enhances the information of different scale features, and improves the performance of the model when dealing with lesions of different sizes. We evaluate our approach on four different datasets and compare it with other methods. Encouragingly, our method achieves significant improvements in performance without losing novelty.

Research on new magnetic epoxy resin composite slurry materials and localization grouting diffusion mechanism

A significant number of steep cracks are frequently encountered in underground engineering, posing a threat to operation. The high-pressure grouting method is a commonly utilized repair technique. Nevertheless, conventional grout is prone to displacement due to its weight, making it challenging to ensure adequate filling of the cracks. Therefore, this study aims to develop a grouting material with targeted displacement and anchoring properties. Firstly, an optimal magnetic slurry composition was determined through an orthogonal test. Subsequently, XRD and SEM were used to analyze the impact of the magnetic field on the composition distribution, internal pore structure, and transient viscosity of the slurry. Afterwards, a model for localized grout diffusion under magnetic was established. The results show that the application of a magnetic field caused the slurry to compact due to magnetic forces, reducing its porosity. Moreover, the dynamic viscosity of the slurry increased exponentially with rising magnetic induction intensity. Notably, a 40.5% increase in the diffusion area was observed when the magnetic field intensity rose from 2500 to 4500 GS. The error between the measured and theoretical values of the magnetic slurry diffusion model was only 8.91%, indicating the model's accuracy in describing the slurry diffusion process under magnetic field influence.

Exploring Solitary Waves and Nonlinear Dynamics in the Fractional Chaffee–Infante Equation: A Study Beyond Conventional Diffusion Models

Exploring Solitary Waves and Nonlinear Dynamics in the Fractional Chaffee–Infante Equation: A Study Beyond Conventional Diffusion Models

Generative diffusion models for synthetic trajectories of heavy and light particles in turbulence

Heavy and light particles are commonly found in many natural phenomena and industrial processes, such as suspensions of bubbles, dust, and droplets in incompressible turbulent flows. Based on a recent machine learning approach using a diffusion model that successfully generated single tracer trajectories in three-dimensional turbulence and passed most statistical benchmarks across time scales, we extend this model to include heavy and light particles. Given the particle type – tracer, light, or heavy – the model can generate synthetic, realistic trajectories with correct fat-tail distributions for acceleration, anomalous power laws, and scale dependent local slope properties. This work paves the way for future exploration of the use of diffusion models to produce high-quality synthetic datasets for different flow configurations, potentially allowing interpolation between different setups and adaptation to new conditions.

Creating High-quality 3D Content by Bridging the Gap Between Text-to-2D and Text-to-3D Generation

In recent times, automatic text-to-3D content creation has made significant progress, driven by the development of pretrained 2D diffusion models. Existing text-to-3D methods typically optimize the 3D representation to ensure that the rendered image aligns well with the given text, as evaluated by the pretrained 2D diffusion model. Nevertheless, a substantial domain gap exists between 2D images and 3D assets, primarily attributed to variations in camera-related attributes and the exclusive presence of foreground objects. Consequently, employing 2D diffusion models directly for optimizing 3D representations may lead to suboptimal outcomes. To address this issue, we present X-Dreamer, a novel approach for high-quality text-to-3D content creation that effectively bridges the gap between text-to-2D and text-to-3D synthesis. The key components of X-Dreamer are two innovative designs: Camera-Guided Low-Rank Adaptation (CG-LoRA) and Attention-Mask Alignment (AMA) Loss. CG-LoRA dynamically incorporates camera information into the pretrained diffusion models by employing camera-dependent generation for trainable parameters. This integration makes the 2D diffusion model camera-sensitive. AMA loss guides the attention map of the pretrained diffusion model using the binary mask of the 3D object, prioritizing the creation of the foreground object. This module ensures that the model focuses on generating accurate and detailed foreground objects. Extensive evaluations demonstrate the effectiveness of our proposed method compared to existing text-to-3D approaches. Our project webpage: https://anonymous-11111.github.io/ . Our code is available at https://github.com/xmu-xiaoma666/X-Dreamer .

Multiphase modeling of pressure-dependent hydrogen diffusivity in fractal porous structures of acrylonitrile butadiene rubber-carbon black composites with different fillers

The hydrogen diffusivities of acrylonitrile butadiene rubber (NBR) composites with different types and contents of carbon black (CB) fillers were investigated in the exposure pressure range of 0.5–10 MPa using a volumetric analysis system. These measured diffusivities exhibited distinct pressure-dependent behavior. The diffusivities of unfilled NBR and NBR composites containing medium-thermal CB filler decreased from 39.7 × 10−11 m2/s to 9.4 × 10−11 m2/s as the pressure increased. On the other hand, the pressure-dependent diffusivities of NBR composites containing high-abrasion furnace, fast-extrusion furnace and semireinforcing furnace CB fillers revealed left-biased unimodal shaped curves, with peak values ranging from 19.7 × 10−11 m2/s to 5.6 × 10−11 m2/s. To model this observed behavior, the diffusion resistance theory with a heterogeneous NBR–CB composite and a fractal porous structure for H2 was introduced. The theoretical parallel diffusion resistance model was found to coexist independently as the surface, Knudsen, and bulk diffusion phases. This theoretical multiphase modeling was applied to the measured diffusivities, determining diffusion resistance parameters for each phase. The obtained individual parametric characteristics for each diffusion were interpreted by considering the CB filler content and volume fraction of the filler. As a result, the diffusivities calculated by multiphase diffusion modeling were in quite agreement with the measured diffusivities for all investigated specimens, where the determined squared correlation coefficient (R2) by fitting process was in the range from 0.69 to 0.97.

Stochastic Resilience in Operations Management Using Ito Diffusion and Gamma Catastrophe Processes

This study introduces a novel Ito diffusion model for operations management, addressing the challenge of maintaining resilience in supply chains and production networks against unpredictable disruptions. The model incorporates a general catastrophe process with a low occurrence rate, using stochastic methods to represent disruption magnitudes as gamma distribution variables. It provides an analytical framework detailing the process's mean, variance, and sample path. Applying this model across various operational scenarios demonstrates its practical significance. By examining the impacts of disruptions on operational efficiency, the model offers insights into disruption dynamics, crucial for resilience planning and risk mitigation. The findings enhance logistics networks' resilience and efficiency, aiding decision-makers in navigating disruptions. This research presents a practical tool for decision-making in operations management and sets the stage for future research with complex variables and emerging technologies to enhance predictive strength in a dynamic environment.

Explicit-implicit priori knowledge-based diffusion model for generative medical image segmentation

The diffusion probabilistic model (DPM) has achieved unparalleled results in current image generation tasks, and some recent research works employed it in several computer vision tasks, such as image super-resolution, object detection, etc. Thanks to DPM's superior ability to generate fine-grained details, these research efforts have yielded significant successes. In this paper, we propose a new DPM-based generative medical image segmentation method, named EIDiffuSeg. Specifically, we first construct an explicit-implicit aggregation priori knowledge with directional supervision ability by mining the semantic distribution pattern in the frequency and spatial domains. Then, the explicit-implicit aggregation priori knowledge is integrated into the different encoding stages of the denoising backbone network using a novel unsupervised priori knowledge induction strategy, which can guide the model to generate a segmentation mask of the region of interest directionally from a random inference process. We evaluate our method on three medical image segmentation benchmark datasets with different modalities and achieve the best segmentation results compared to state-of-the-art methods. Especially, compared to several current diffusion-based image segmentation methods, we achieved a 9% Dice improvement in the polyp segmentation benchmark. Our code will be available at https://github.com/Notmezhan/EIDiffuSeg.