87 results about "Linear modulation" patented technology

The invention provides an industrial defect image generation system and method based on deep learning, and relates to the technical field of artificial intelligence. A defect form adaptive module, a physical attribute modulation module and a multi-mechanism fusion module are integrated in a defect image generation module; dynamically selecting a feature extraction unit according to the defect type label based on a pre-constructed generative network model, and generating a defect feature map; generating an affine transformation parameter based on the physical attribute vector, and performing channel-by-channel linear modulation on a defect feature map of a middle layer of the generative network model, so that a multi-scale defect feature map finally generated by the generative network model contains specified defect type features and physical attribute features; and fusing the multi-scale defect feature image and the defect-free background image to obtain a fused defect image. The problems that in the prior art, industrial defect image generation is insufficient in sense of reality, poor in controllability and poor in fusion effect are solved, the method can be used for data enhancement of industrial visual inspection, and downstream model performance is improved.

The invention discloses an intelligent agent digital image interaction generation method based on multi-modal perception, which comprises the following steps: collecting multi-modal input data of a user, and respectively carrying out preprocessing and feature extraction on the multi-modal input data; inputting to an improved efficient modal cross learning network, and carrying out multi-modal feature fusion processing; constructing a semantic intention map, introducing a time index edge weight and an emotion driving edge weight, and encoding the map by using a structure perception map neural network; a modal style vector is extracted through a cross-modal style contrast learning mechanism, and a personalized style coding vector is generated through a hierarchical nested structure; inputting a personalized regulation and control gating mechanism, and regulating and controlling the middle layer representation in the interaction strategy generation process by adopting a feature channel linear modulation method; inputting the representation vector into a behavior strategy generation module to generate a multi-modal behavior output sequence; and the sequence is output to drive the digital image to perform synchronous response, and natural response generation in the user interaction process is completed.

The invention discloses a long-tail data generation method based on feed-forward stylized reconstruction. The method comprises the following steps: constructing multi-modal geometric perception input and feature codes; reconstructing a semantic guidance scene based on a feature-level linear modulation mechanism; the invention relates to 3D consistency controllable style migration based on a feedforward reconstruction model. According to the invention, a feedforward reconstruction network based on DINOv2 improvement is constructed. Multi-modal adaptation is performed on a backbone network input layer, and decoupled high-level semantic features are explicitly injected by using a feature-level linear modulation mechanism, so that the geometric accuracy of automatic driving scene reconstruction is improved. On the basis of a three-dimensional scene of feedforward reconstruction, a distillation strategy is adopted, and the generation capability of a 2D diffusion model is migrated to a 3D Gaussian field. According to the method, the phenomena of object deformation, disappearance and the like in the generated data are effectively avoided, and long-tail weather data with geometric consistency of rain, snow and the like can be generated through the text instruction.

The invention discloses a small sample crack recognition method and system based on transfer learning, and the method comprises the steps: constructing an innovative and deep-coupled neural network architecture, and enabling a self-adaptive crack perception attention module integrated with a parallel asymmetric convolution kernel to recognize the linear geometric features of a crack, and generating an attention graph; then, the attention map is adopted to carry out pixel-by-pixel signal pre-modulation on an original input image, the enhanced image is sent to a pre-training MobileNetV4 backbone network integrated with a feature-level linear modulation layer, and channel-level dynamic adaptation of feature flow in the network is realized by learning affine transformation parameters; and finally, the extracted depth features are sent to a Gaussian naive Bayes classifier for classification, and depth geometric analysis is carried out on the attention map so as to realize interpretable fracture severity evaluation. According to the method, the accuracy, robustness and interpretability of crack identification are remarkably improved, and the technical bottleneck in a small sample scene is effectively solved.

The application discloses a kind of based on multiple exposure generation fusion's low-light image enhancement system, belong to low-light image enhancement technical field.System includes: adaptive multiple exposure adjustment module (AMEA), joint prediction gamma coefficient, linear modulation factor and compensation parameter, to input low-light image carries out nonlinear and linear combination transformation, generates multiple luminance, contrast and different color response multiple exposure candidate image;Multiple exposure fusion module (MEF), original low-light image and candidate image channel splicing are carried out, and preliminary enhancement image is obtained by fine-grained weighting fusion of channel fusion coefficient and introduction of perception compensation parameter correction;Color refinement module (CRM), using encoder decoder structure and introducing fast fourier representation in bottleneck layer, realize spatial domain and frequency domain joint modeling, carry out global color calibration and local detail optimization to preliminary enhancement image.The overall network uses lightweight design such as depth separable convolution and dense connection, inference is simple, and can be run in real time in edge device.The application generates by adaptive multiple exposure, structure constraint fusion and frequency domain refinement, effectively restores dark texture, suppresses over-enhancement and color deviation, improves visual naturalness and robustness in complex low-light scene.

This invention discloses a method for inverting sea surface wind speed from GNSS reflected signals based on mask perception and image-based Mamba-CNN. The method includes: performing spatiotemporal matching and interpolation processing on multi-source spaceborne GNSS-R observation data and ERA5 reanalysis data; gridding the processed observation parameters according to geographical location to obtain a multi-channel grid image and acquiring a binarized occupancy mask; inputting the multi-channel grid image and the binarized occupancy mask into a mask perception Mamba-CNN model; calibrating the grid features through a mask conditional encoder and a feature linear modulation layer, and performing spatial modeling using feature extraction blocks combining convolutional and Mamba state space paths to obtain the wind speed inversion result; aggregating the extracted features using center-perception pooling, and outputting the final sea surface wind speed inversion image through a regression network. This invention can significantly reduce computational complexity, effectively reconstruct the continuous wind field structure, and improve the inversion accuracy and robustness in high-wind-speed areas.

Provided in the embodiments of the present application are a speech enhancement method, a speech enhancement model training method and a computer program product. The speech enhancement method comprises: acquiring a speech data sample of a target speaker, and performing spectral feature extraction on the speech data sample on the basis of a bidirectional long short-term memory network layer and an attention statistics pooling layer, so as to obtain a spectral feature of the target speaker; inputting the spectral feature into a feature-wise linear modulation layer in a time-frequency narrow-band neural network, and by means of the feature-wise linear modulation layer and on the basis of the spectral feature, adjusting a spectral feature corresponding to mixed speech data extracted by the time-frequency narrow-band neural network, wherein the mixed speech data at least comprises speech data of a plurality of speakers including the target speaker; and on the basis of the adjusted spectral feature, performing speech enhancement processing on the speech data of the target speaker in the mixed speech data, so as to obtain the enhanced speech data of the target speaker.

The present invention relates to a control system (20) and method for controlling the flow of current between a voltage source (22) and a multiphase motor or other application / load for controlling a multiphase motor (28). The control system includes, between each phase of the multiphase motor and the voltage source: a high-side diode (30A-C) and a high-side switch (34A-C) positioned in parallel between a positive terminal of the voltage source and the motor, wherein the high-side diode is reverse biased with respect to the positive terminal; and a low-side diode (32A-C) and a low-side switch (36A-C) positioned in parallel between a negative terminal of the voltage source and the motor, wherein the low-side diode is reverse biased with respect to the motor. The control system operates the high-side and low-side switches to provide efficient operation of the motor.

The present application relates to the technical field of image processing, and more particularly to a lesion segmentation method of adaptive dynamic text prompt, comprising: acquiring a lesion image, and generating an adaptive dynamic text prompt based on the lesion image; constructing a multi-modal enhanced fusion text prompt adapter based on a FiLM global channel recalibration module, a spatial cross-attention interaction module and a zero initialization gate nonlinear integration module cascade; the global channel recalibration module of FiLM utilizes a feature linear modulation mechanism to perform channel-level weighting on a visual feature map; before the feature enters spatial interaction, according to the text semantic enhancement band response, the activation of the background noise channel is inhibited. The present application overcomes the problems of insufficient utilization of multi-modal prompts, dependence on artificial static labeling of text prompts, serious dependence on prior information in the reasoning stage and difficulty in multi-modal feature fusion in a low-contrast environment in the existing medical image segmentation technology.

This invention provides a production monitoring system for mirror cabinet processing equipment based on multi-source data, relating to the field of process control. The system acquires multi-source data from the mirror cabinet through an acquisition module and constructs a discrete mesh model; a deviation topology construction module generates directional selective edge weights and a nonlinear modulation function; a propagation control coupling module achieves synchronous iterative convergence of deviation states and control components; a topology constraint module identifies structural boundaries and divides control sub-regions; a region control module generates region control vectors, which are then decomposed into processing adjustment quantities and coupled in by the instruction module; and a feedback reconstruction module optimizes parameters based on the processing results. This invention enables deviation propagation to match the processing path, achieving bidirectional coupling of state and control, thus improving accuracy and control stability in the processing of complex curved surfaces and multi-material mirror cabinets.

This application discloses a soft-start method, apparatus, device, and storage medium for a three-phase LLC converter. The soft-start method includes: responding to a start command, pre-charging the three phases of the three-phase LLC converter in a preset sequence; after the three phases have completed pre-charging, performing coordinated linear modulation on the duty cycle and frequency of the three phases so that the duty cycle of each phase approaches a preset target duty cycle, and the frequency of each phase approaches a preset target frequency; and controlling the three-phase regulated output when the duty cycle of each phase reaches the preset target duty cycle and the frequency of each phase reaches the preset target frequency. This application can achieve more stable and safer three-phase soft starting.

The invention discloses a motor linear overmodulation method and device, a vehicle and a storage medium, and the method comprises the steps: obtaining a reference voltage component and a DC bus voltage of a motor, and obtaining a reference voltage amplitude and a reference voltage phase angle according to the reference voltage component; a modulation coefficient is obtained according to the reference voltage amplitude and the direct-current bus voltage, the working area of the motor is judged according to the modulation coefficient, and the working area is divided into a linear modulation area, an overmodulation I area and an overmodulation II area according to the modulation coefficient; and when the working area is the overmodulation II area, obtaining a dynamic weight coefficient according to the reference voltage phase angle, and controlling the motor according to the dynamic weight coefficient. Therefore, when the working area of the motor is the overmodulation II area, the motor is controlled according to the dynamic weight coefficient obtained by the voltage phase angle, so that the voltage jump of the overmodulation II area is eliminated, the power performance of the electric vehicle driving system is improved to the maximum extent, and meanwhile, the output smoothness is ensured.

The invention discloses a direct-current-bias-free high-power LED visible light signal transmission method and system, and relates to the technical field of optical communication, and the method comprises the steps: carrying out 64QAM-OFDM modulation and IQ modulation on original audio and video data, carrying out AC coupling peak-to-peak value extraction and DC component filtering, and then carrying out single-power-supply amplification and pre-equalization; and a driving positive voltage signal is generated to drive a high-power LED to emit light, and an optical communication signal is generated. Through a low-loss alternating-current coupling transmission mode, the modulated real signal peak-to-peak value is subjected to coupling transmission and single positive power supply amplification, so that the signal is changed into positive voltage to control the LED light intensity and form an LED optical communication signal, direct-current voltage does not need to be introduced into a digital domain, the problem of bandwidth reduction is avoided, and the signal transmission efficiency is improved. The direct-current bias voltage of the LED can be kept stable, the LED works in a linear modulation area, bandwidth expansion is carried out through the pre-equalization circuit, and the functions of LED illumination and high-speed communication can be considered.

A method (100) for determining a power dissipation value of an electronic power circuit during an electrical period of a modulated control signal is disclosed, comprising: determining a linear mean switching energy value (110) of the electronic power circuit based on a modulated control signal with linear modulation; determining a nonlinear mean switching energy value (120) of the electronic power circuit based on a modulated control signal with nonlinear modulation, in particular overmodulation; determining a scaling factor value (130) based on the linear mean switching energy value and the nonlinear mean switching energy value; and determining the power dissipation value (140) based on the linear mean switching energy value and the scaling factor value.Furthermore, a computing unit designed and equipped to execute the procedure (100) is disclosed.

The application discloses a panoramic image depth estimation method combining spherical correction and semi-supervised learning, innovatively designs a spherical correction (SR) mechanism based on spherical coordinate Fourier coding and feature linear modulation (FiLM), a lightweight plug-and-play PanoAdapter adaptive module, a teacher-student semi-supervised cross-domain knowledge transfer framework and a joint optimization strategy fusing pseudo-label distillation and consistency constraint, realizes apparent correction of equidistant cylindrical projection (ERP) nonlinear spherical distortion, alleviates the contradiction between panoramic labeling data scarcity and perspective model cross-domain migration difficulty, and improves the geometric consistency and generalization ability of panoramic depth estimation in diversified real scenes.

This invention discloses a method for preparing epitaxial stress-controlled polarized metals, belonging to the field of functional oxide thin film preparation. The method includes: performing a special high-temperature annealing pretreatment on a YAlO3(111) substrate; calibrating the pulsed laser deposition rate and using a moving mask technique to prepare a wedge-shaped thin film with continuously varying thickness; and precisely controlling key process parameter windows such as growth temperature, laser energy density, oxygen partial pressure, and the distance between the target and the substrate. This invention obtains an ideal nucleation template by optimizing the substrate pretreatment process, achieves linear modulation of the thickness gradient by precisely controlling the matching of the mask moving speed and deposition rate, and obtains a high-quality polarized metal thin film with a room temperature resistivity as low as 173 μΩ·cm and stability across the entire temperature range by systematically optimizing key process parameters. This method has the advantages of high repeatability, high control precision, and wide applicability, and can be extended to the preparation of other perovskite oxide thin films, showing broad application prospects in the field of polarized electronic devices.

The invention relates to the technical field of industrial measurement, and discloses a collaborative automatic measurement system for industrial field multivariable features, comprising: an acquisition module for generating a sampling data stream with a synchronous timestamp generated by a hardware synchronous clock; the wave velocity calculation module is used for extracting the real-time deviation amount of the main variable sampling value relative to the reference value and determining the real-time wave velocity of the propagation medium by using a first-order linear modulation equation; the dynamic addressing module is used for calculating the propagation lag time length of the auxiliary variable relative to the main variable, and backtracking in the historical cache queue according to the propagation lag time length so as to extract the auxiliary variable historical sampling value with the aligned time sequence; according to the method, time axis accumulated drift errors caused by violent evolution of propagation medium attributes can be countered, and it is ensured that the system maintains extremely high measurement fidelity and physical stability under the full-dynamic working condition.

The invention discloses an RGB-D sight line tracking method fusing head postures, and belongs to the field of computer sight line estimation. Facial feature extraction adopts a RetinaFace network as a detection core, a detected human face region is cropped through an affine transformation strategy, a cropped image is input into a 3D Dense Face Alliance Version 2 (3DDFAV2) model after being normalized, and an eye RGB-D image block and a calculated head posture Euler angle are output through registration. In order to effectively integrate complementary information of RGB and a depth mode, a lightweight multi-mode sight tracking network is provided and is divided into three main levels: a feature extraction layer, a feature fusion layer and a regression prediction layer. According to the network, a double-branch multi-mode framework is constructed, five input modes including a binocular RGB image, a depth map and a head posture are processed respectively, Kalman filtering smoothing is utilized, a multi-mode information fusion strategy combining an attention and feature linear modulation (FiLM) mechanism is designed, and finally high-precision sight line regression is achieved through a multi-layer perceptron (MLP).

This invention provides a motion environment perception method based on biomechanical features and gait adaptation, belonging to the field of human-computer interaction perception. The method includes: acquiring the subject's original motion feature sequence and biomechanical feature vector; scaling and aligning the physical dimensions using a biomechanical scaling matrix to obtain a motion feature tensor; inputting the motion feature tensor and biomechanical feature vector into a multi-task temporal convolutional network for feature-level linear modulation to obtain modulated deep features; processing these features through a backbone network, with the classification branch outputting terrain categories and the regression branch outputting initial values ​​of environmental geometric parameters; freezing the backbone network parameters; fine-tuning the regression branch based on an adaptive loss function driven by gait phase and foot arch features to obtain a motion environment perception model; inputting real-time data from the subject; and outputting terrain categories and environmental geometric parameters. This invention solves the technical problem of low accuracy in environmental geometric parameter estimation caused by biomechanical differences between individual subjects in existing technologies.

The application provides an electromagnetic electrostatic double-drive Coppe filter chip and a preparation method thereof, and relates to the fields of optoelectronic chips and semiconductor technologies.The filter chip comprises a movable mirror body, a fixed mirror body, a permanent magnet, a PCB circuit board and a spacing layer.A first reflecting mirror, a first detection electrode, a second detection electrode, a first drive electrode and a second drive electrode are arranged on the end face of the movable mirror body.The first detection electrode and the first drive electrode are arranged on one side of the first reflecting mirror, and the second detection electrode and the second drive electrode are arranged on the other side of the first reflecting mirror.A permanent magnet slot for embedding the permanent magnet is arranged on the other end face of the movable mirror body.The application can overcome the problems of the single-drive Coppe filter chip, such as nonlinear modulation, limited tuning range and poor monochromaticity.

The invention discloses a Transform runoff prediction method and system fused with an enhanced interannual climate memory mechanism, runoff prediction is carried out based on a runoff prediction network model of an encoder and a decoder of a Transform, and the method comprises the steps of obtaining dynamic data and static data of a to-be-studied drainage basin, carrying out continuous short-term window and long-term window division on the preprocessed dynamic data, and carrying out runoff prediction on the to-be-studied drainage basin. Forming a short-term dynamic input feature and a long-term dynamic input feature; compressing the static feature into a global vector by adopting an encoder, performing linear modulation on the global vector and the short-term dynamic input feature, and performing feature extraction to obtain a short-term feature sequence; carrying out sparse sampling on the long-term dynamic input features by adopting an encoder, and carrying out feature extraction to obtain a long-term feature sequence; fusing the short-term feature sequence and the long-term feature sequence by using an encoder to obtain a fusion feature of bidirectional information interaction; the decoder predicts runoff prediction features according to the fusion features, and a softplus activation function of the decoder is adopted for adjustment to obtain the runoff prediction features.

The application discloses a washing machine motor operation control method and system based on fundamental wave voltage control, and relates to the technical field of washing machine motor control. The method comprises the following steps: S1, offline calculation of a relationship curve between the fundamental wave voltage and the reference voltage in the over-modulation region and storage of the curve in the memory; S2, calculation of the fundamental wave voltage by using a field-oriented control algorithm, wherein the d-axis and q-axis components of the fundamental wave voltage and the stator voltage are obtained through a speed loop and a current loop; S3, if the fundamental wave voltage is in the linear modulation region or exceeds the over-modulation region, an SVPWM algorithm in a 120-degree coordinate system is used for modulation; if the fundamental wave voltage is in the over-modulation region, an over-modulation algorithm in the 120-degree coordinate system is used to calculate the three-phase voltage action time, and finally, the SVPWM algorithm is used for modulation. The application directly takes the fundamental wave voltage as a control target, avoids the uncertainty caused by the inconsistency between the reference voltage and the fundamental wave voltage, and thus realizes accurate control of the fundamental wave voltage which has an actual effect.

The invention discloses an audio deep forgery detection method and device, equipment, a storage medium and a product, and the method comprises the steps: carrying out the detection of to-be-processed audio data through a pre-trained audio deep forgery detection model, and carrying out the detection of the to-be-processed audio data in the detection process of the audio deep forgery detection model, extracting audio frequency features and audio time features of the audio data based on a time-frequency double-branch structure, and enhancing breathing sound clues in the audio time features through a breathing sound feature linear modulation mechanism; performing cross attention fusion on the audio time features and the audio frequency features, and predicting the fused features to obtain a forged audio detection result; according to the method, the breathing sound feature linear modulation mechanism and the cross attention fusion mechanism are utilized, breathing sound clue enhancement and fusion of the breathing sound clues and the time-frequency features can be achieved, natural voice fine clues such as breathing sound are effectively recognized, and therefore the accuracy of audio deep forgery detection is improved.

The application discloses a two-dimensional ferroelectric transistor device based on controllable nucleation to realize linear modulation and a preparation and micro-characterization method thereof, and the method comprises the following steps: sequentially preparing a bottom electrode and a ferroelectric layer on a substrate; preparing a planar structure covering the surface of the ferroelectric layer, wherein the planar structure comprises a conductive layer and a two-dimensional semiconductor layer, and is used for adjusting the internal electric field distribution of the ferroelectric layer to control the nucleation of domains to occur at the interface between the conductive layer and the two-dimensional semiconductor layer; preparing an insulating layer and a metal electrode to complete the preparation of the two-dimensional ferroelectric transistor device; and micro-characterizing the two-dimensional ferroelectric transistor device to verify the domain nucleation and flipping rule and obtain the device resistance state change under different domain distributions. The application realizes linear flipping of ferroelectric domains by introducing a controllable nucleation process, prepares a transistor device with linear conductance modulation characteristics, and micro-verify by using advanced micro-characterization technology, and provides a closed-loop process which can guide process optimization.

Embodiments of the present application disclose an audio separation method and device, electronic equipment and storage medium, which can reduce the amount of calculation in audio separation by encoding the sample audio spectrum information, thereby improving the efficiency of audio separation; the linear feature vector can be obtained by linear modulation of the sample audio feature vector, and the linear feature vector of the initial bottleneck layer feature vector of the replacement bottleneck layer is decoded and processed, the parameters in the bottleneck layer can be updated, thereby improving the robustness of the deep learning segmentation network; training the deep learning segmentation network according to the target loss value can improve the accuracy of the deep learning segmentation network in audio separation; when the deep learning segmentation network is used for audio separation of the audio to be separated, no matter how short the length of the audio to be separated is, and no matter what kind of financial business audio is, the effect of audio separation can also be achieved, that is, the accuracy of audio separation of the audio to be separated can be improved.

The invention relates to the technical field of microseismic signal processing, in particular to a self-adaptive microseismic signal noise reduction method. The method comprises the following steps: constructing a conditional denoising model for adaptively denoising a microseismic signal, fusing a U-Net infrastructure, a Transform module and a characteristic linear modulation module conditional modulation mechanism, constructing a paired data set, preprocessing the data set, then verifying the denoising performance of a CDU-Net model through a training and evaluation process, and optimizing the denoising performance of the CDU-Net model, so that the denoising performance of the CDU-Net model is improved. And explainable evidence is provided for the rationality of a network architecture through visual analysis, and the contribution degree of each component in the model is analyzed through an ablation experiment, so that the microseismic signal noise reduction model which can adapt to the change of a complex geological environment and effectively suppress non-stationary noise is realized through the mode, and the noise reduction efficiency of the microseismic signal is improved. The problem of self-adaptive noise reduction of micro-seismic signals in a complex geological environment is solved.

The invention discloses a permanent magnet synchronous motor phase current reconstruction method and device for reducing harmonic waves, and belongs to the technical field of motor control. According to the method, a sampling topological structure is optimized, a sampling resistor is shifted from a direct-current bus side to a position between upper bridge arms of two phases A and B of an inverter, and low-harmonic and high-precision phase current reconstruction is realized by combining a novel switch state phase shift method (N switch state phase shift strategy) and a dynamic zero vector distribution strategy. The method specifically comprises the steps that sectors are divided according to a space vector diagram, and sampling vectors are selected; phase shift operation is carried out in a low modulation degree area through minimization of PWM waveform modification, and the average voltage conservation principle is followed; zero vector time is dynamically allocated in a high modulation degree region to extend a linear modulation range. The phase current total harmonic distortion (THD) is remarkably reduced, the modulation range is expanded, meanwhile, the cost effectiveness and the real-time performance are considered, and the method is suitable for low-cost permanent magnet synchronous motor driving systems such as electric automobile fans and water pumps.

The invention provides a multi-mode linearly-modulated electromagnetic structure which comprises a metamaterial base body, a plurality of channels arranged at intervals are arranged in the thickness direction of the metamaterial base body, and controllable adjusting media are arranged in the channels. According to the multi-mode electromagnetic structure capable of being linearly modulated, geometric reconstruction of an electromagnetic response structure of the structure is completed by adjusting the content of a medium in a channel, and then the transmission and cut-off states of the structure are changed. The invention provides a multi-mode linearly-modulated electromagnetic structure, a method of injecting / discharging liquid in a channel in a segmented and partitioned manner is adopted, compared with a traditional diode loading method, a new design thought is introduced, and linear adjustment of electromagnetic wave transmission, reflection and absorption capacity is facilitated.

This invention discloses a lightweight multi-task context control image processing system, including a first image encoder, a second image encoder, a multi-scale fusion module, and an output feature processing module. The multi-scale fusion module includes N multi-scale fusion units connected sequentially from high to low spatial resolution. The first and second image encoders are used to extract features and downsample images from sample paths and control paths. The multi-scale fusion units are used to fuse query sample features and task control features at different spatial resolutions. Each multi-scale fusion unit includes a residual convolution module, a sliding window cross-attention module, a feature linear modulation module, a refinement module, and a downsampling module. This invention uses lightweight modules such as depthwise separable convolution and group normalization, significantly reducing the number of parameters compared to traditional control networks. The output feature channel is compatible with downstream diffusion models and can be plugged into existing diffusion models without extensive modifications to the original model architecture.