White Noise Research Articles

White noise’s effect on premature infants’ physiological parameters during peripheral intravenous catheter insertion

BackgroundEarly and continuous exposure to painful stimuli in premature infants leads to short-and long-term complications. Listening to white noise is an accessible and inexpensive non-invasive method that can be used as a safe nursing intervention in hospitals. This study aimed to assess white noise’s effect on premature Infants’ physiological parameters during peripheral intravenous catheter insertion.MethodsThe present experimental study was conducted on 40 premature Infants. From 5 min before Indwelling catheters to 10 min after, white noise was played through headphones to infants at a controlled volume in the test group. Using a monitor, physiological parameters were recorded from 10 min before to 30 min after the Indwelling catheter. Statistical analysis was done through the SPSS version 27 software program.ResultsRegarding respiratory rate, there was no statistically significant difference between the two groups in the first, second, and sixth stages (p < 0.05). However, in the third, fourth, and fifth stages, RR in the control group was significantly higher than the test group (p ≤ 0.05). In terms of mean arterial blood pressure, there was no statistically significant difference between the two groups in any of the stages (p < 0.05). In terms of heart rate, there was no statistically significant difference between the two groups in the first, second, and sixth stages (p < 0.05). However, in the third, fourth, and fifth stages, HR in the control group was significantly higher than the test group (p ≤ 0.05). Regarding oxygen saturation percentage, there was no statistically significant difference between the two groups in the first, second, third, and sixth stages (p < 0.05). In contrast, in the fourth and fifth stages, the oxygen saturation percentage in the test group was significantly higher than in the control group (p < 0.05).ConclusionsAlthough in both groups the painful procedure led to an increase in physiological variables (respiratory rate, heart rate, mean arterial blood pressure) and a decrease in blood oxygen saturation, in the test group these variables approached their initial state more quickly within half an hour after exposure to white noise. This confirms that the effect of white noise on the improvement of physiological variables is gradual.Clinical trial numberNot applicable.

Establishment and extension of a compact and robust binary feature descriptor for UAV image matching

Abstract Local feature description is a crucial challenge in unmanned aerial vehicle (UAV) images matching. Although scholars have explored a variety of methods for defining local feature descriptions, improving the matching accuracy of UAV images and reducing the memory consumption of descriptor remain worthwhile research issues. A compact and robust binary feature descriptor for UAV image matching is proposed. To achieve rotational invariance of the descriptor, a sampling pattern based on the region overlap error is employed to assign the dominant direction of the feature points. Additionally, a rotation invariant gradient computation method is proposed to reduce the dominant direction assignment errors. Then, the region around the feature point is divided into grids based on the region overlap error, and gradient orientation histogram and average intensity histogram are constructed. Finally, these histograms are binarized to generate a 30-byte binary descriptor. Results from the Oxford dataset, the Iguazu dataset, and a high-resolution UAV image dataset indicate that the proposed descriptor is compact and achieves the highest matching accuracy compared to nine commonly used feature descriptors, including scale-invariant feature transform, speeded up robust feature, binary robust invariant scalable keypoints, KAZE, oriented fast and rotated brief, fast retina keypoint, local intensity order pattern, boosted efficient binary local image descriptor, and triplet-based efficient binary local image descriptor. Notably, on the high-resolution UAV images with translational transformations, scale variations, salt-and-pepper noise, and Gaussian white noise, the accuracies of the proposed descriptor were 87.09%, 80.83%, 92.46%, and 87.42%, respectively. The improvements ranged from 6.95% to 53.20%, 9.32% to 55.90%, 10.39% to 44.29%, and 6.54% to 58.24%.

Adversarial purification using a modified guided diffusion model and adaptive self-supervised training

Abstract As the applications of deep neural networks broaden, enhancing their robustness against adversarial attacks becomes increasingly important. Multiple studies have demonstrated their vulnerability to adversarial perturbations. This paper presents an iterative method employing a probabilistic diffusion model, guided by a self-supervised training strategy, to purify potential adversarial inputs. Specifically, the diffusion process blends the adversarial noise with incrementally added Gaussian noise. Subsequently, both types of noise are removed during the guided denoising process. Unlike existing methods that use a fixed number of iterations and rely on the adversarial input for guidance, our scoring-based approach dynamically adjusts the purification duration for each individual image, thereby reducing computational overhead and minimizing the side effects caused by excessive purification. Additionally, we modify the input at each iteration using a self-supervised strategy and then utilize this modified input to guide the denoising process, resulting in better adversarial robustness. Since the proposed method operates without any label information, it can be applied to various training paradigms, including supervised, semi-supervised, self-supervised, and unsupervised learning. We conduct several experiments on widely used machine vision datasets to evaluate the efficacy of the proposed method. The results confirm that our method effectively eliminates adversarial perturbations across these datasets. For example, under a white-box PGD attack with an l ∞ ball (ϵ = 8/255) on CIFAR-10, our method achieves a robust accuracy of 92.13%, surpassing the state-of-the-art by 3.68%.

Sensitive hydrogen detection using off-axis integrated cavity output spectroscopy at 2.1 μm

Reliable hydrogen (H2) detection is essential for advancing hydrogen energy applications. In this study, we report a sensitive H2 detection method using off-axis integrated cavity output spectroscopy. A distributed-feedback laser with fiber output was used to target the H2 absorption transition near 2122 nm. The optical cavity was constructed using two spherical mirrors, each with a reflectivity of 99.985%, providing an effective absorption path length of 2500 m. The Galatry line shape function was selected to accurately fit the narrow absorption feature of H2. Furthermore, we achieved a 1.74-fold improvement in the signal-to-noise ratio by applying radio frequency white noise perturbations. The system's performance was demonstrated by measuring H2 mixtures across a broad concentration range from 1040 ppm to 20.24%. Finally, the Allan–Werle deviation analysis revealed a detection limit of 53 ppm H2 at an average time of 400 s.

Multi-Patch Time Series Transformer for Robust Bearing Fault Detection with Varying Noise

In time-series studies involving bearing sensor data, Gaussian noise and white noise techniques are commonly employed to evaluate model robustness. However, these conventional noise techniques are limited in their applicability to real-world industrial environments. This paper proposes three novel noise techniques—electrical interference noise, harmonic noise, and random shock noise—that more accurately reflect the complex noise encountered in industrial settings. Additionally, a new deep learning model, MultiPatchTST, is introduced, demonstrating robust performance under various noise conditions. Experimental results reveal that Gaussian noise has minimal impact on model performance, whereas the proposed noise techniques significantly affect performance, providing a more realistic evaluation of noise robustness. The proposed MultiPatchTST model achieves superior performance across all metrics in the presence of all four noise types, confirming its robustness and reliability.

Adversarial Robustness for Deep Learning-Based Wildfire Prediction Models

Rapidly growing wildfires have recently devastated societal assets, exposing a critical need for early warning systems to expedite relief efforts. Smoke detection using camera-based Deep Neural Networks (DNNs) offers a promising solution for wildfire prediction. However, the rarity of smoke across time and space limits training data, raising model overfitting and bias concerns. Current DNNs, primarily Convolutional Neural Networks (CNNs) and transformers, complicate robustness evaluation due to architectural differences. To address these challenges, we introduce WARP (Wildfire Adversarial Robustness Procedure), the first model-agnostic framework for evaluating wildfire detection models’ adversarial robustness. WARP addresses inherent limitations in data diversity by generating adversarial examples through image-global and -local perturbations. Global and local attacks superimpose Gaussian noise and PNG patches onto image inputs, respectively; this suits both CNNs and transformers while generating realistic adversarial scenarios. Using WARP, we assessed real-time CNNs and Transformers, uncovering key vulnerabilities. At times, transformers exhibited over 70% precision degradation under global attacks, while both models generally struggled to differentiate cloud-like PNG patches from real smoke during local attacks. To enhance model robustness, we proposed four wildfire-oriented data augmentation techniques based on WARP’s methodology and results, which diversify smoke image data and improve model precision and robustness. These advancements represent a substantial step toward developing a reliable early wildfire warning system, which may be our first safeguard against wildfire destruction.

Distributed Consensus Gossip-Based Data Fusion for Suppressing Incorrect Sensor Readings in Wireless Sensor Networks

Incorrect sensor readings can cause serious problems in Wireless Sensor Networks (WSNs), potentially disrupting the operation of the entire system. As shown in the literature, they can arise from various reasons; therefore, addressing this issue has been a significant challenge for the scientific community over the past few decades. In this paper, we examine the applicability of seven distributed consensus gossip-based algorithms for sensor fusion (namely, the Randomized Gossip algorithm, the Geographic Gossip algorithm, three initial configurations of the Broadcast Gossip algorithm, the Push-Sum protocol, and the Push-Pull protocol) to compensate for incorrect data in WSNs. More specifically, we consider a scenario where the sensor-measured data (measured by a set of independent sensor nodes) are skewed due to Gaussian noise with a various standard deviation σ, resulting in discrepancies between the measured values and the true value of observed physical quantities. Subsequently, the aforementioned algorithms are employed to mitigate this skewness in order to improve the accuracy of the measured data. In this paper, WSNs are modeled as random geometric graphs with various connectivity, and the performance of the algorithms is evaluated using two metrics (specifically, the mean square error (MSE) and the number of sent messages required for an algorithm to be completed). Based on the presented results, it is identified that all the examined algorithms can significantly suppress incorrect sensor readings (MSE without sensor fusion = −0.42 dB if σ = 1, and MSE without sensor fusion = 14.05 dB if σ = 5), and the best performance is achieved by PS in dense graphs and by GG in sparse graphs (both algorithms achieve the maximum precision MSE = −24.87 dB if σ = 1 and MSE = −21.02 dB if σ = 5). Additionally, the performance of the analyzed distributed consensus gossip algorithms is compared to the best deterministic consensus algorithm applied for the same purpose.

Perturbations of optical solitons in magneto-optic waveguides incorporating multiplicative white noise and sixth-order dispersion: a study of the Sasa–Satsuma equation

Perturbations of optical solitons in magneto-optic waveguides incorporating multiplicative white noise and sixth-order dispersion: a study of the Sasa–Satsuma equation

Data-Driven Identification of Variational Equations for Vortex-Induced Vibration Systems

Abstract In this study, a data-driven approach using the embedded variational principle is used to identify the variational equations of vortex-induced vibration fluid–structure interaction systems, in particular the coupling term and the aerodynamic damping term. Under the data-driven paradigm, variational equation identification is primarily accomplished through five steps: collecting discrete data, setting variational functions, building the product function, solving linear equations, and evaluating errors. The explicit variational equations of the system are eventually determined automatically from the excitation and response. Gaussian white noise is added to the excitation to evaluate the method's noise robustness. The findings demonstrate that numerical estimation which stays away from higher-order derivatives significantly enhances the variational law identification's noise robustness by taking advantage of the variational law's lower-order time derivatives. Furthermore, the arbitrariness of the variational setting inherent in the variational law significantly improves the effectiveness of data utilization and lowers the necessary data volume. In addition, a system of linear equations is solved by identifying connected nonlinear equations, which significantly increases modeling efficiency. The basis for engineering modeling, optimization, and control of intricate fluid–structure interaction systems are provided by these benefits.

Classification of Flying Drones Using Millimeter-Wave Radar: Comparative Analysis of Algorithms Under Noisy Conditions

This study evaluates different machine learning algorithms in detecting and identifying drones using radar data from a 60 GHz millimeter-wave sensor. These signals were collected from a bionic bird and two drones, namely DJI Mavic and DJI Phantom 3 Pro, which were represented in complex form to preserve amplitude and phase information. The first benchmarks used four algorithms, namely long short-term memory (LSTM), gated recurrent unit (GRU), one-dimensional convolutional neural network (Conv1D), and Transformer, and they were benchmarked for robustness under noisy conditions, including artificial noise types like white noise, Pareto noise, impulsive noise, and multipath interference. As expected, Transformer outperformed other algorithms in terms of accuracy, even on noisy data; however, in certain noise contexts, particularly Pareto noise, it showed weaknesses. For this purpose, we propose Multimodal Transformer, which incorporates more statistical features—skewness and kurtosis—in addition to amplitude and phase data. This resulted in a improvement in detection accuracy, even under difficult noise conditions. Our results demonstrate the importance of noise in processing radar signals and the benefits afforded by a multimodal presentation of data in detecting unmanned aerial vehicle and birds. This study sets up a benchmark for state-of-the-art machine learning methodologies for radar-based detection systems, providing valuable insight into methods of increasing the robustness of algorithms to environmental noise.

Conditional diffusion-generated super-resolution for myocardial perfusion MRI

IntroductionMyocardial perfusion MRI is important for diagnosing coronary artery disease, but current clinical methods face challenges in balancing spatial resolution, temporal resolution, and slice coverage. Achieving broader slice coverage and higher temporal resolution is essential for accurately detecting abnormalities across different slice locations but remains difficult due to constraints in acquisition speed and heart rate variability. While techniques like parallel imaging and compressed sensing have significantly advanced perfusion imaging, they still suffer from noise amplification, residual artifacts, and potential temporal blurring due to the rapid transit of dynamic contrast vs. the temporal constraints of the reconstruction.MethodsThis study introduces a conditional diffusion-based generative model for myocardial perfusion MRI super resolution, addressing the trade-offs between spatiotemporal resolution and slice coverage. We adapted Denoising Diffusion Probabilistic Models (DDPM) to enhance low-resolution perfusion images into high-resolution outputs without requiring temporal regularization. The forward diffusion process introduces Gaussian noise incrementally, while the reverse process employs a U-Net architecture to progressively denoise the images, conditioned on the low-resolution input image.ResultsWe trained and validated the model on a retrospective dataset of dynamic contrast-enhanced (DCE) perfusion MRI, consisting of both stress and rest images from 47 patients with heart disease. Our results showed significant image quality improvements, with a 5.1% reduction in nRMSE, a 1.1% increase in PSNR, and a 2.2% boost in SSIM compared to GAN-based super-resolution method (P &amp;lt; 0.05 for all metrics using paired t-test) in retrospective study. For the 9 prospective subjects, we achieved a total nominal acceleration of 8.5-fold across 5–6 slices through a combination of low-resolution acquisition and GRAPPA. PerfGen outperformed GAN-based approach in sharpness (4.36 ± 0.38 vs. 4.89 ± 0.22) and overall image quality (4.14 ± 0.28 vs. 4.89 ± 0.22), as assessed by two experts in a blinded evaluation (P &amp;lt; 0.05) in prospective study.DiscussionThis work demonstrates the capability of diffusion-based generative models in generating high-resolution myocardial perfusion MRI from conditional low-resolution images. This approach has shown the potentials to accelerate myocardial perfusion MRI while enhancing slice coverage and temporal resolution, offering a promising alternative to existing methods.

Non-invasive quantification of bone (re) modeling dynamics in adults with osteogenesis imperfecta treated with Setrusumab using timelapse HR-pQCT.

Timelapse imaging using high-resolution peripheral quantitative computed tomography (HR-pQCT) has emerged as a noninvasive method to quantify bone (re)modelling. However, there is no consensus on how to perform the procedure. As part of the ASTEROID phase-2b multicenter trial, we used 29 same-day repeated scans from adults with osteogenesis imperfecta (OI) to identify a method that minimized measurement error. We evaluated input image type, registration method, segmentation mask, and for grayscale images various values for the voxel density difference considered formed or resorbed, minimum formation/resorption cluster size, and Gaussian smoothing sigma. We verified the accuracy of our method and then used it on longitudinal scans (baseline, 6, 12, 18 and 24month) from 78 participants to assess bone formation and resorption induced by an anabolic (setrusumab) and anti-catabolic (zoledronic acid) treatments as part of the ASTEROID trial. Regardless of image registration method, binary input images resulted in large errors ~13% and ~ 8% for first- and second-generation scanners, respectively. For the grayscale input images, errors were smaller for 3D compared to matched angle registration. For both scanner generations, a density threshold of 200mgHA/cm3 combined with Gaussian noise reduction resulted in errors <1%. We verified the method was accurate by showing that similar regions of bone formation and resorption were identified when comparing each scan from the same-day repeated scans with a scan from another timepoint. Timelapse analysis revealed a dose-dependent increase in bone formation and resorption with setrusumab treatment. Zoledronic acid altered bone changes in favor of formation, although no changes reached statistical significance. This study identifies a timelapse method that minimizes measurement error, which can be used in future studies to improve the uniformity of results. This noninvasive imaging biomarker revealed dose dependent bone (re) modeling outcomes from one year of setrusumab treatment in adults with OI.

Precision of the acoustic control of single-photon scattering with semiconductor quantum dots

Acoustic modulation of quantum dots (QDs) allows one to control the scattering of photons. Here we theoretically characterize the degree of this acoustic control in the frequency domain. We formulate the theory of low-intensity resonance fluorescence (RF) in the presence of white noise and show that a high level of control is achievable with a two-tone acoustic field for appropriate settings of modulation amplitudes as long as the noise-induced phase diffusion coefficient remains one order of magnitude smaller than the acoustic frequency. In addition, using a quantitative model of optical signal collection, we determine that the acoustic phase must be stable over 104 to 105 acoustic periods for efficient control.

Stability of a delayed prey–predator model with fear, stochastic effects and Beddington–DeAngelis functional response

In view of the importance of predator-dependent functional response and fear of prey induced by powerful predators, we construct a delayed prey–predator model with fear and Beddington–DeAngelis functional response. The existence, uniqueness, and global asymptotic stability of equilibrium points are investigated and some criteria are established. Next, Hopf bifurcation analysis is executed, and the critical values of such bifurcation parameters as fear and delay for the determinate system are obtained. Then we extend it to a random environment and study the boundedness of expectation of solutions and the global asymptotic stability. Finally, the main findings are validated by numerical examples. It is worth noting that the specific influences of fear by predator, time delay, and white noise are explored numerically. Simulation figures intuitively exhibit that fear, delay, and white noise bring serious influences on the stability of the system. Fear from predator leads to a lower equilibrium state of prey and predator, and it can change the system stability from unstable to stable after exceeding a certain critical value. The time delay has a significant impact on the system stability by producing Hopf bifurcations accompanied by limit cycles, and even lead to multiple stabilities. Larger white noise can change the system stability from stable to unstable.

Background white noise increases neuronal activity by reducing membrane fluctuations and slow-wave oscillations in auditory cortex.

The brain faces the challenging task of preserving a consistent portrayal of the external world in the face of disruptive sensory inputs. What alterations occur in sensory representation amidst noise, and how does brain activity adapt to it? Although it has previously been shown that background white noise (WN) decreases responses to salient sounds, a mechanistic understanding of the brain processes responsible for such changes is lacking. We investigated the effect of background WN on neuronal spiking activity, membrane potential, and network oscillations in the mouse central auditory system. We found that, in addition to increasing background spiking activity in the auditory cortex and thalamus, background WN decreases neural activity fluctuations, as reflected in the membrane potential of single neurons and the local field potential. Blocking acetylcholine signaling in the auditory cortex eliminated the WN-dependent increase in background activity as well as the shift in slow-wave oscillations. Together, our observations show that background WN is not filtered away along the auditory pathway, but rather drives sustained changes in cortical activity that can be reverted by blocking cholinergic inputs.

Frequency-Modulated Antipodal Chaos Shift Keying Chaotic Communication on Field Program Gate Array: Prototype Design and Performance Insights

Using chaos for communication can provide more robust channel security, covert transmission, and inherent support for spread-spectrum modulation. Although numerous studies have explored this technology, its practical deployment remains limited due to substantial hardware demands, complex signal processing, and a lack of efficient modulation methods for chaotic signals. In this study, a novel chaotic digital communication system is proposed and studied. A prototype of a frequency-modulated antipodal chaos shift keying (FM-ACSK) system is implemented on an Intel Cyclone V field-programmable gate array (FPGA) along with a complete mathematical model using Matlab R2022a Simulink software. Using FPGAs to implement chaotic oscillators avoids analog system problems such as component drift and high thermal instability while providing determined system parameters, rapid prototyping, and high throughput. The employment of FM over a chaotic modulation layer provides a passband operation (currently at an intermediate frequency of 10.7 MHz) while adding the benefits of carrier frequency offset robustness and constant signal envelope. Within this study, the robustness of FM-ACSK to white noise in the channel was evaluated using bit error rate, which was tested through hardware experiments and simulations. The results show the feasibility and potential performance limitations of this approach to chaotic communication system design.

White noise induces transitions between dynamical states in a network of nonlocally coupled birhythmic van der Pol-type systems

White noise induces transitions between dynamical states in a network of nonlocally coupled birhythmic van der Pol-type systems

Prediction Accuracy of Stackelberg Game Model of Electricity Price in Smart Grid Power Market Environment

With the deepening of power market reform and the increasingly fierce competition in the power market, the accurate prediction of electricity price has become an important demand for power market participants to make scientific decisions, optimize resource allocation, and reduce risks. Electricity price forecast can provide a key reference for the power market, help market participants make wise decisions, promote competition and efficient operation and cope with complex market fluctuations, provide a scientific basis for various entities to optimize resource allocation, reduce risks and improve benefits, and promote the sustainable development of the power industry. This study presents a dynamic retail price prediction method for smart grid based on the Stackelberg game model. Firstly, the correlation test is used to verify the strong correlation between electric load and electricity price. Secondly, the parameters of the Stackelberg model are determined, and the load and electricity price are tested using the white noise test. Finally, by comparing the BP neural network model and quantifying the model parameters, the superiority of the model is verified. The results show that the Stackelberg game model has higher prediction accuracy than the BP neural network model in electricity price prediction.

YOLOv8‐ESW: An Improved Oncomelania hupensis Detection Model

ABSTRACTTraditional Oncomelania hupensis detection relies on human eye observation, which results in reduced efficiency due to easy fatigue of the human eye and limited individual cognition, an improved YOLOv8 O. hupensis detection algorithm, YOLOv8‐ESW(expectation–maximization attention [EMA], Small Target Detection Layer, and Wise‐IoU), is proposed. The original dataset is augmented using the OpenCV library. To imitate image blur caused by motion jitter, salt and pepper, and Gaussian noise were added to the dataset; to imitate images from different angles captured by the camera in an instant, affine, translation, flip, and other transformations were performed on the original data, resulting in a total of 6000 images after data enhancement. Considering the insufficient feature fusion problem caused by lightweight convolution, We present the expectation–EMA module (E), which innovatively incorporates a coordinate attention mechanism and convolutional layers to introduce a specialized layer for small target detection (S). This design significantly improves the network's ability to synergize information from both superficial and deeper layers, better focusing on small target O. hupensis and occluded O. hupensis. To tackle the challenge of quality imbalance among O. hupensis samples, we employ the Wise‐IoU (WIoU) loss function (W). This approach uses a gradient gain distribution strategy and improves the model convergence speed and regression accuracy. The YOLOv8‐ESW model, with 16.8 million parameters and requiring 98.4 GFLOPS for computations, achieved a mAP of 92.74% when tested on the O. hupensis dataset, marking a 4.09% improvement over the baseline model. Comprehensive testing confirms the enhanced network's efficacy, significantly elevating O. hupensis detection precision, minimizing both missed and false detections, and fulfilling real‐time processing criteria. Compared with the current mainstream models, it has certain advantages in detection accuracy and has reference value for subsequent research in actual detection.

BTSegDiff: Brain tumor segmentation based on multimodal MRI Dynamically guided diffusion probability model.

In the treatment of brain tumors, accurate diagnosis and treatment heavily rely on reliable brain tumor segmentation, where multimodal Magnetic Resonance Imaging (MRI) plays a pivotal role by providing valuable complementary information. This integration significantly enhances the performance of brain tumor segmentation. However, due to the uneven grayscale distribution, irregular shapes, and significant size variations in brain tumor images, this task remains highly challenging. In order to overcome these obstacles, we have introduced a novel framework for automated segmentation of brain tumors that leverages the diverse information from multi-modal MRI scans. Our proposed method is named BTSegDiff and it is based on a Diffusion Probability Model (DPM). First, we designed a dynamic conditional guidance module consisting of an encoder. This encoder is used to extract information from multimodal MRI images and guide the DPM in generating accurate and realistic segmentation masks. During the guidance process, we need to fuse the diffused generated features with the extracted multimodal features. However, diffusion process itself introduces a significant amount of Gaussian noise, which can affect the fusion results. Therefore, we designed a Fourier domain feature fusion module to transfer this fusion process to Euclidean space and reduce the impact of high-frequency noise on fusion. Lastly, we have taken into account that the DPM, as a generative model, produces non-unique results with each sampling. In the meticulous field of medicine, this is highly detrimental. Therefore, we have designed a Stepwise Uncertainty Sampling module based on Monte Carlo uncertainty calculation to generate unique outcomes and enhance segmentation accuracy simultaneously. To validate the effectiveness of our approach, we perform a validation on the popular BraTs2020 and BraTS2021 benchmarks. The experimental results show that our method outperforms many existing brain tumor segmentation methods. Our code is available at https://github.com/jaceqin/BTSegDiff.