Abstract: This disclosure provides systems, devices, apparatus, and methods, including computer programs encoded on storage media, for variable rate tessellation. A graphics processor may receive data for geometry processing of a plurality of patches in a draw call. The graphics processor may reduce a tessellation factor of each of the plurality of patches based on a property of each of the plurality of patches. The reduced tessellation factor may correspond to a TRF. The property may correspond to a shading rate or a number of visible pixels. The graphics processor may apply the TRF for each of the plurality of patches. The graphics processor may render each of the plurality of patches based on the applied TRF for each of the plurality of patches.

Abstract: The various embodiments described herein include methods, devices, and systems for generating object meshes. In some embodiments, a method includes obtaining a trained classifier, and an input observation of a 3D object. The method further includes generating a three-pole signed distance field from the input observation using the trained classifier. The method also includes generating an output mesh of the 3D object from the three-pole signed distance field; and generating a display of the 3D object from the output mesh.

Abstract: Methods of and systems for characterization of a 3D point cloud are disclosed. The method comprises accessing a 3D point cloud, the 3D point cloud being a set of data points representative of the object, determining, based on the 3D point cloud, a 3D reconstructed object, determining, based on the 3D reconstructed object, a digital framework of the 3D point cloud, the digital framework being a ramified 3D tree structure, the digital framework being representative of a base structure of the object, morphing a 3D reference model of the object onto the 3D reconstructed object, the morphing being based on the digital framework; and determining, based on the morphed 3D reference model and the 3D reconstructed object, characteristics of the object.

Abstract: Systems and methods are provided for determining a location of a selection in a space viewable in a camera view on a display of a computing device, detecting movement of the computing device, and generating a path based on the location of the selection and the movement of the computing device. The systems and methods further provide for generating a three-dimensional (3D) mesh along the path, populating the 3D mesh with selected options to generate a 3D paint object, and causing the generated 3D paint object to be displayed. The systems and methods further provide for receiving a request to send a message comprising an image or video overlaid by the 3D paint object, capturing the image or video overlaid by the displayed 3D paint object, generating the message comprising the image or video overlaid by the 3D paint object, and sending the message to another computing device.

Abstract: Methods and systems are disclosed for performing operations comprising: receiving a video that includes a depiction of a real-world object; generating a three-dimensional (3D) body mesh associated with the real-world object that tracks movement of the real-world object across frames of the video; obtaining an external mesh associated with an augmented reality element; automatically establishing a correspondence between the 3D body mesh associated with the real-world object and the external mesh; deforming the external mesh based on movement of the real-world object and the established correspondence with the 3D body mesh; and modifying the video to include a display of the augmented reality element based on the deformed external mesh.

Abstract: Provided is a method for generating high-quality isosurface mesh in real time, which takes the Marching Cubes algorithm as a baseline to efficiently generate a high-quality mesh of a 3D model, re-examines the case table in the MC algorithm, and puts forward the concept of equivalent edges. By combining with the remeshing technology, the MC algorithm is optimized from three aspects: deleting equivalent edges with the worst performance from the case table by using connectivity modification and vertex insertion technology; changing the geometric shape of the active edge to make it more perpendicular to the isosurface; and moving the shared cut points of cube cells. According to the present application, the mesh with a higher quality is generated at the running speed close to the standard MC algorithm, and the mesh quality is close to the post-processing remeshing algorithm with extremely high time consumption.

Abstract: Systems and methods include determination of an entropy value associated with each of a plurality of voxels of a three-dimensional image, determination, for each of the plurality of voxels, of a respective filter based on the entropy value associated with the voxel, wherein a first filter determined for a first voxel associated with a first entropy value is different from a second filter determined for a second voxel associated with a second entropy value different from the first entropy value, application, for each of the plurality of voxels, of the respective filter to a value of the voxel to generate a replacement value for the voxel, and generation of a filtered three-dimensional image based on the generated replacement value of each of the plurality of voxels.

Abstract: In accordance with some embodiments, an exemplary process for initializing members of a shared enhanced reality setting is described. In accordance with some embodiments, an exemplary process for forming a private sub-space and example features of the private sub-space is described.

Abstract: A productivity tool for an analytical laboratory is presented. The productivity tool provides aggregated operational status of all laboratory instruments within an analytical laboratory, even without access to their respective internal data stream (i.e., legacy and third party instruments) by determining operational status of a first group of instruments by capturing and processing internal data streams comprising data indicative of an operational status. The productivity tool also determines operational status of a second group of instruments by capturing and processing external surveillance stream(s) of one or more laboratory instrument(s) of the second group.

Abstract: Virtual resources can be controlled from within an augmented reality environment. For example, a computing device can display, in an augmented reality environment, a representation of a physical object corresponding to a virtual resource. The computing device can receive an interaction with the representation of the physical object in the augmented reality environment. The computing device can output a command to cause a result in the virtual resource external to the augmented reality environment based on the interaction.

Abstract: An ambulation simulating apparatus, including: a user feet interface comprising foot rests; a dynamic inertia mechanism configured to provide an inertial load when a downwardly force is applied due to the foot and a kinematic connection configured to enable transmission of motion from the feet interface to the dynamic inertia and backwards as the dynamic inertia mechanism resists said downwardly applied force by the foot.

Abstract: There is provided an information processing device, an information processing method, and a program capable of appropriately displaying an object in a virtual space. Included is a setting unit that sets an initial position at which an object is initially displayed on a display unit on the basis of presentation position information regarding a presentation position to be presented in a virtual three-dimensional space of the object superimposed and displayed on a real world, size information regarding a size of the object, and field of view area information regarding a field of view area of the display unit that displays the object, in which the setting unit sets the initial position on a deeper side in a depth direction than a position represented by the presentation position information in a coordinate system based on the real world. The present technology can be applied to, for example, an information processing device that provides augmented reality (AR).

Abstract: Methods, devices, and systems related to a computing device for capturing an augmented reality (AR) image displaying an AR are described. An example method can include projecting a structured array of light, from a mobile device, onto an object. The method further includes measuring multi-directional sections of the object with the projected light. The method further includes reconstructing a three-dimensional (3D) figure of the object based on the measured multi-directional sections. The method further includes displaying an augmented reality (AR) image associated with the 3D figure on a user interface of the mobile device.

Abstract: Augmented reality presentations are provided at respective electronic devices. A first electronic device receives information relating to modification made to an augmented reality presentation at a second electronic device, and the first electronic device modifies the first augmented reality presentation in response to the information.

Abstract: Aspects of the present disclosure involve a system comprising a computer-readable storage medium storing at least one program and method for performing operations comprising: receiving, by a messaging application, a video feed from a camera of a user device that depicts a face; receiving a request to add a 3D caption to the video feed; identifying a graphical element that is associated with context of the 3D caption; and displaying the 3D caption and the identified graphical element in the video feed at a position in 3D space of the video feed proximate to the face depicted in the video feed.

Abstract: Systems and methods include transforming a digital file into a three-dimensional object that is spatially positioned in a three-dimensional virtual environment to visually organize the digital file relative to the three-dimensional virtual environment. Embodiments of the present disclosure relate to receiving the digital file that includes digital file parameters and is in a file format. The digital file is transforming into the three-dimensional object based on the digital file parameters associated with the digital file. The three-dimensional object is representative of the presentation of the digital file when executed by the computing device. The three-dimensional object is spatially positioned at a spatial location in the three-dimensional environment based on the digital file parameters of the digital file.

Abstract: The present technology is directed to presenting a 3-D representation of an object, which is captured in a 2-D image of the object in an environment in a 3-D simulation of the environment. The present technology can receive 3-D data representing the environment including image data of the object in the environment, and a label identifying the object. The present technology can further locate a position of the object in the 3-D simulation of the environment based on determining a depth and an orientation of the object in the environment based on a semantic database of attributes associated with the object, obtain a 3-D representation of the object from the semantic database of attributes associated with the object, and project the 3-D representation of the object into the 3-D simulation of the environment at the determined position for the object.

Abstract: A stereoscopic image acquisition method includes acquiring a first three-dimensional model corresponding to a two-dimensional image according to the two-dimensional image and a depth image; acquiring a target area and an area to be repaired corresponding to the first three-dimensional model; obtaining a second three-dimensional model by repairing a pixel in the area to be repaired according to a color value and a depth value of a pixel in the target area; and acquiring a stereoscopic image to be displayed according to a preset viewing angle change path and the second three-dimensional model.

Abstract: An information processing system, a program, and an information processing method that can improve convenience when operating a movable object within an application. An information processing system that provides an application that can be executed on an information processing terminal includes a detection unit that detects a three-dimensional motion of a finger of a user, and a motion control unit that controls a three-dimensional motion of a movable object within the application based on the three-dimensional motion of the finger of the user detected by the detection unit.

Abstract: Various implementations disclosed herein include devices, systems, and methods that generate a combined 3D representation of a user based on an alignment based on a 3D reference point. For example, a process may include obtaining a predetermined three-dimensional (3D) representation that is associated with a 3D reference point defined relative to a skeletal representation of the user. The process may further include obtaining a sequence of frame-specific 3D representations corresponding to multiple instants in a period of time, each of the frame-specific 3D representations representing a second portion of the user at a respective instant of the multiple instants in the period of time. The process may further include generating combined 3D representations of the user generated by combining the predetermined 3D representation with a respective frame-specific 3D representation based on an alignment which is based on the 3D reference point.

Abstract: Systems and methods are disclosed for aligning a two-dimensional (2D) design image to a 2D projection image of a three-dimensional (3D) design model. One method comprises receiving a 2D design document, the 2D design document comprising a 2D design image, and receiving a 3D design file comprising a 3D design model, the 3D design model comprising one or more design elements. The method further comprises generating a 2D projection image based on the 3D design model, the 2D projection image comprising a representation of at least a portion of the one or more design elements, generating a projection barcode based on the 2D projection image, and generating a drawing barcode based on the 2D design image. The method further comprises aligning the 2D projection image and the 2D design image by comparing the projection barcode and the drawing barcode.

Abstract: Artefacts in a sequence of image frames may be reduced or eliminated through modification of an input image frame to match another image frame in the sequence, such as by geometrically warping to generate a corrected image frame with a field of view matched to another frame in sequence of frames with the image frame. The warping may be performed based on a model generated from data regarding the multi-sensor device. The disparity between image frames may be modeled based on image captures from the first and second image sensor for scenes at varying depths. The model may be used to predict disparity values for captured images, and those predicted disparity values used to reduce artefacts resulting from image sensor switching. The predicted disparity values may be used in image conditions resulting in erroneous actual disparity values.

Abstract: A system for capturing, organizing, and storing handwritten notes includes a plurality of boundary markers. The boundary markers are configured to be positioned on a writing surface. The system also includes a tangible non-transitory computer readable medium encoded with instructions which, when run on a camera-equipped computing device, causes the camera-equipped computing device to execute processes. The processes include capturing an image of the writing surface with the markers thereon. The processes also include detecting the boundary markers in the captured image. Additionally, the processes include identifying a virtual boundary in the captured image based on the positions of the boundary markers. The processes then unwarp a portion of the captured image within the virtual boundary to produce an unwarped image.

Abstract: Disclosed are an image processing method and device, a training method of a neural network and a storage medium. The image processing method includes: obtaining an input image, and processing the input image by using a generative network to generate an output image. The generate network includes a first sub-network and at least one second sub-network, and the processing the input image by using the generative network to generate the output image includes, processing the input image by using the first sub-network to obtain a plurality of first feature images; performing a branching process and a weight sharing process on the plurality of first feature images by using the at least one second sub-network to obtain a plurality of second feature images; and processing the plurality of second feature images to obtain the output image.

Abstract: There is included a method and apparatus comprising computer code configured to cause a processor or processors to perform obtaining an input low resolution (LR) image comprising a height, a width, and a number of channels, implementing a feature learning deep neural network (DNN) configured to compute a feature tensor based on the input LR image, generating, by an upscaling DNN, a high resolution (HR) image, having a higher resolution than the input LR image, based on the feature tensor computed by the feature learning DNN, wherein a networking structure of the upscaling DNN differs depending on different scale factors, and wherein a networking structure of the feature learning DNN is a same structure for each of the different scale factors.

Abstract: In order to more accurately white balance an image, weightings can be determined for pixels of an image when computing an illuminant color value of the image and/or a scene. The weightings can be based at least in part on the Signal-to-Noise Ratio (SNR) of the pixels. The SNR may be actual SNR or SNR estimated from brightness levels of the pixels. SNR weighting (e.g., SNR adjustment) may reduce the effect of pixels with high noise on the computed illuminant color value. For example, one or more channel values of the illuminant color value can be determined based on the weightings and color values of the pixels. One or more color gain values can be determined based on the one or more channel values of the illuminant color value and used to white balance the image.

Abstract: An image inpainting method is performed at a computing device, the method including: displaying an original image, and determining a to-be-inpainted first original region and a second original region different from the first original region in the original image in accordance with a user input; identifying a target pixel and a target pixel block comprising the target pixel in the first original region; obtaining a target search region from the second original region according to the target pixel and the target pixel block comprising the target pixel, an area of the target search region being smaller than an area of the second original region; and searching the target search region for a reference pixel block matching the target pixel block and updating the original image by inpainting the target pixel block according to the reference pixel block, and displaying the updated image.

Abstract: Data processing systems (e.g. image processing systems) and methods are provided for processing a stream of data values (e.g. pixel values). The image processing system comprises a processing module configured to: receive a plurality of pixel values; and implement processing of a particular pixel value by operating on a particular subset of the received pixel values, by: defining a set of one or more groups into which pixel values within the particular subset can be grouped; classifying each of the pixel values within the particular subset into one of the groups of the set of one or more groups based on the value of that pixel value; processing the particular pixel value using one or more of the pixel values of the particular subset in dependence on the classification of the pixel values of the particular subset into the one or more groups; and outputting the processed particular pixel value.

Abstract: Provided are a method and apparatus for presenting image for a VR device, a device and a non-transitory computer-readable storage medium. The method for presenting image for a VR device includes steps of: acquiring input image data; establishing a correspondence between the input image data and output image data in a polar coordinate system, the polar coordinate system having a pole which is an intersection of a center of a field of view of the VR device and a display plane; obtaining image data to be presented according to the output image data; and presenting the image data to be presented on the basis of a convex lens.

Abstract: An image correction method and a processor are disclosed. The method includes performing a feature point search on a quick response (QR) code image to determine multiple feature points, dividing a coded area of the QR code image into multiple sub-regions according to the multiple feature points, determining a compensation vector for each sub-region according to the feature points corresponding to each sub-region, and compensating and correcting each sub-region according to the compensation vector of each sub-region to obtain a corrected image. Thus, the solution provided by the present application can avoid interference between different sub-regions by means of correcting the QR code image in a regional manner using the compensation vectors, thereby more accurately correcting the distortion of the QR code image.

Abstract: Various improvements to video processing are described, including highlight recovery by dividing an image into different frequency ranges and compressing the lower range before recombining, tone mapping to avoid hue shifts, dynamic tone mapping based on detected scene changes in the video, dynamic tone mapping using the shape of a tone mapping curve that is changed based on the histogram of each video frame, HDR Luminance Channel Repair of artifacts, changing the shape of the tone mapping curve based on a histogram of each video frame, upscaling Chroma by using Luma channel information motion interpolation using Neural Networks, motion compensated noise reduction using Neural Networks to filter random noise and film grain, grain/noise agnostic upscaling using Neural Networks, and hiding video frames by strategically dropping or repeating them in unnoticeable places to prevent a visible video stutter based on scene characteristics to accommodate clock variations.

Abstract: An ultrasound imaging system has an improved dynamic range control which enables a user to reduce image haze with substantially no effect on bright tissue in the image and without increasing speckle variance. A dynamic range processor processes an input image to produce an approximation image which is a spatially low pass filtered version of the input image. The dynamic range of the approximation image is compressed, and image detail, unaffected by the compression, is added back to produce a dynamically compressed image for display.

Abstract: Image analysis and processing may include a first sensor readout unit receiving first Bayer format image data corresponding to a first input image, a second sensor readout unit receiving second Bayer format image data corresponding to a second input image, a first Bayer-to-RGB unit obtaining first RGB format image data based on the first Bayer format image data, a second Bayer-to-RGB unit obtaining second RGB format image data based on the second Bayer format image data, a high dynamic range unit configured obtaining high dynamic range image data based on a combination of the first RGB format image data and the second RGB format image data, an RGB-to-YUV unit obtaining YUV format image data based on the high dynamic range image data, and a three-dimensional noise reduction unit obtaining processed image data based on the YUV format image data.

Abstract: Disclosed herein are system, apparatus, article of manufacture, method and/or computer program product embodiments, and/or combinations and sub-combinations thereof, for dynamic tone mapping of video content. An example embodiment operates by identifying, by a dynamic tone mapping system executing on a media device, characteristics of a first video signal having a first dynamic range based on a frame-by-frame analysis of the first video signal. The example embodiment further operates by modifying, by the dynamic tone mapping system, a tone mapping curve based on the characteristics of the first video signal to generate a modified tone mapping curve. Subsequently, the example embodiment operates by converting, by the dynamic tone mapping system, the first video signal based on the modified tone mapping curve to generate a second video signal having a second dynamic range that is less than the first dynamic range.

Abstract: An image processing method includes: (a) receiving a to-be-processed image; (b) obtaining a target brightness of the to-be-processed image; (c) creating a histogram curve of the to-be-processed image; (d) clipping the histogram curve according to a clipping value; (e) obtaining a mapping function of the clipped histogram curve; (f) obtaining a mapped image mapped according to the mapping function; (g) obtaining an average brightness of the mapped image; (h) obtaining a difference between the target brightness and the average brightness; and (i) if the difference is greater than a threshold, adjusting the clipping value and repeating steps (d) to (i) until the difference is less than the threshold. Thus, if the to-be-processed image does not reach the target brightness, the image processing device automatically repeats steps (c) to (i) until the to-be-processed image reaches the target brightness.

Abstract: A method and a system for processing an image and transform it into a high resolution and high-definition image using a computationally efficient image transformation procedure is provided. The transformation of the image comprises first transforming the image, also referred to as an intensity image, into a layered distance field (DF) image comprising an ordered sequence of multiple layers. Each layer in the ordered sequence is associated with a DF procedure and a set of rules for mapping the DF values to intensity values of the respective layer. The result of applying the DF procedures to each location in the intensity image is a transformed intensity image, which is of high definition and high resolution. The application of the DF procedures is governed by a stopping criteria based on error values between the intensity image and a reconstructed intensity image.

Abstract: A computer-implemented method to perform image-to-image translation. The method can include obtaining one or more machine-learned generator models. The one or more machine-learned generator models can be configured to receive an input image and a user-specified conditioning vector that parameterizes one or more desired values for one or more defined characteristics of an output image. The one or more machine-learned generator models can be configured to perform, based at least in part on the user-specified conditioning vector, one or more transformations on the input image to generate the output image with the one or more desired values for the one or more defined characteristics. The method can include receiving the input image and the user-specified conditioning vector. The method can include generating, using the machine-learned generator model, an output image having the one or more desired values for the one or more characteristics.

Abstract: The scaled Peak Signal-to-Noise Ratio (PSNR) method scales PSNR to 1.00 or 100% from infinite decibel values. The PSNR describes the ratio between the maximum possible value (power) of a signal and the power of signal's noise. Noise affects the quality of its representation. Signals dynamically fluctuate from the smallest to largest possible quantity. The unsettled range to maximum from minimum values of the noise is unknown. Therefore, the PSNR usually forms on a logarithmic way (log) of describing a ratio, the decibel (dB). The dB is able to gauge the level of power, communication signal (intensity or voltage), sound pressure, and intensity of several other items including images.

Abstract: An image processing method implemented by a processor includes receiving an image, acquiring a target image that includes an object from the image, calculating an evaluation score by evaluating a quality of the target image, and detecting the object from the target image based on the evaluation score.

Abstract: The present disclosure provides a visual model for image analysis of material characterization and analysis method thereof. By collecting and labeling big data of microscopic images, the present disclosure establishes an image data set of material characterization; and uses this data set for high-throughput deep learning, establishes a neural network model and dynamic statistical model based on deep learning, to identify and locate atomic or lattice defects, and automatically mark the lattice spacing, obtain the classification and statistics of the true shape of the microscopic particles of the material, quantitatively analyze the tissue dynamics of the material.

Abstract: In an abnormality detection device for detecting whether there is an abnormality in either of an optical sensor and a camera that are used to detect objects around a vehicle, a correlation value calculation unit calculates a correlation value between at least one of a reflection intensity image that is detected by the optical sensor emitting light toward surroundings of the vehicle and then receiving reflected light of emitted light and a background light image that is detected by the optical sensor receiving light from the surroundings of the vehicle while the optical sensor is emitting no light, and a captured image of the surroundings from the camera. An abnormality determination unit is configured to, in response to the correlation value calculated by the correlation value calculation unit being out of a predefined range, determine that there is an abnormality in either of the optical sensor and the camera.

Abstract: A fault detection method and system for tunnel dome lights based on an improved localization loss function. The method includes: constructing a dataset of tunnel dome light detection images; acquiring a you only look once (YOLO) v5s neural network based on the improved localization loss function; training the YOLO v5s neural network according to the dataset to obtain a trained YOLO v5s neural network; acquiring a to-be-detected tunnel dome light image; detecting, with the trained YOLO v5s neural network, the to-be-detected tunnel dome light image to obtain position coordinates of luminous dome lights; and determining, according to the position coordinates of the luminous dome lights, whether a fault occurs in the tunnel dome lights. The present disclosure can accurately localize the tunnel dome lights and label the positions, and can detect whether the tunnel dome lights work normally according to a relative positional relationship between the labeled dome lights.

Abstract: Provided are a method and a device for real-time monitoring of tunnel deformation. The monitoring method includes: S1, erecting a plurality of structured light sources in an unstable area to be monitored of a tunnel structure, and erecting a monitoring terminal in a relatively stable region of the tunnel structure, where the monitoring terminal may communicate with the plurality of structured light sources; S2, observing all structured lights in the unstable area of the tunnel structure and obtaining structured light curves in real time by the monitoring terminal; S3, analyzing imaging changes of the structured lights, detecting deformation degrees and offset distances of the tunnel in real time, and monitoring the diseases such as settlement, convergence of a single-section of the tunnel and integral settlement of a multi-section of the tunnel by the data processing unit.

Abstract: In an illustrative embodiment, a system for identifying products on a production line includes image capturing devices that acquire images of containers moving along a production line at an inspection location. The system also includes a rejection device and a controller that configures the image capturing devices for image acquisition based on properties of the containers, identifies a product associated with each of the containers based on a portion of a product identification code and a portion of additional features detected in the images, and determines whether the identified product matches predetermined properties or characteristics, resulting in a pass result, otherwise a non-pass result occurs. When a non-pass result occurs, the controller outputs a signal to actuate the rejection device that removes the container from the production line.

Abstract: An object inspection apparatus according to an embodiment disclosed herein includes: a first flipper apparatus configured to rotate a first object; a second flipper apparatus configured to rotate a second object; and a single camera device configured to move from a position corresponding to one of the first flipper apparatus and the second flipper apparatus to a position corresponding to the other, the camera device being configured to inspect object surfaces of the first object and the second object. Each of the first flipper apparatus and the second flipper apparatus includes at least one flipper unit.

Abstract: Provided is a pavement nondestructive detection and identification method based on small samples, including: constructing an original dataset, dividing the original dataset into several patch blocks, sampling the patch blocks, and obtaining samples of the patch blocks; inputting the samples of the patch blocks into a Transformer model for feature extraction and target reconstruction, and obtaining a trained Transformer model; and based on the trained Transformer model, detecting input pavement sample images.

Abstract: Surgical kit inspection systems and methods are provided for inspecting surgical kits having parts of different types. The surgical kit inspection system comprises a vision unit including a first camera unit and a second camera unit to capture images of parts of a first type and a second type in each kit and to capture images of loose parts from each kit that are placed on a light surface. A robot supports the vision unit to move the first and second camera units relative to the parts in each surgical kit. One or more controllers obtain unique inspection instructions for each of the surgical kits to control inspection of each of the surgical kits and control movement of the robot and the vision unit accordingly to provide output indicating inspection results for each of the surgical kits.

Abstract: A method of controlling the quality of printed products by using a computer includes always producing multiple printed products with at least partially identical image content in a printing machine in a course of a print job to be completed. The multiple printed products are recorded by at least one image sensor and are sent to the computer as digital image data. The computer examines and assesses the digital image data in an image inspection process to find print defects and to sort out printed products that have been found to be unusable. The computer assesses the identical image content of the digital image data of at least two consecutive printed products and only takes into consideration such detected print defects that are present on the at least two assessed consecutive printed products.

Abstract: The present disclosure discloses an Internet of Things (IoT) system for industrial data processing, a control method, and a storage medium, and provides a technical solution for a timely adjustment of production line parameters according to image information generated during a machine vision data collection on the production line. Through a difference of the image information corresponding to different process operations, a processing situation of different process operations may be obtained, so that more accurate adjustment of the production line parameter may be achieved without increasing the system complexity, thereby effectively reducing a development cost of the IoT, and increasing accuracy of the intelligent manufacturing control.

Abstract: Systems and methods process images to determine a skin condition severity analysis and to visualize a skin analysis such as using a deep neural network (e.g. a convolutional neural network) where a problem was formulated as a regression task with integer-only labels. Auxiliary classification tasks (for example, comprising gender and ethnicity predictions) are introduced to improve performance. Scoring and other image processing techniques may be used (e.g. in assoc. with the model) to visualize results such as highlighting the analyzed image. It is demonstrated that the visualization of results, which highlight skin condition affected areas, can also provide perspicuous explanations for the model. A plurality (k) of data augmentations may be made to a source image to yield k augmented images for processing. Activation masks (e.g. heatmaps) produced from processing the k augmented images are used to define a final map to visualize the skin analysis.