Abstract: Generalizable feature distillation systems that align 3D features with 2D foundation model features using a feedforward network, avoiding per-scene optimization, and a flexible end-to-end 3D scene interpretation system that applies the extracted 3D features and pretrained 2D vision-language models for various 3D scene understanding tasks.

Abstract: In various embodiments, a training application trains a machine learning model to generate three-dimensional (3D) representations of two-dimensional images. The training application maps a depth image and a viewpoint to signed distance function (SDF) values associated with 3D query points. The training application maps a red, blue, and green (RGB) image to radiance values associated with the 3DI query points. The training application computes a red, blue, green, and depth (RGBD) reconstruction loss based on at least the SDF values and the radiance values. The training application modifies at least one of a pre-trained geometry encoder, a pre-trained geometry decoder, an untrained texture encoder, or an untrained texture decoder based on the RGBD reconstruction loss to generate a trained machine learning model that generates 3D representations of RGBD images.

Abstract: In various embodiments, a scene reconstruction model generates three-dimensional (3D) representations of scenes. The scene reconstruction model computes a first 3D feature grid based on a set of red, blue, green, and depth (RGBD) images associated with a first scene. The scene reconstruction model maps the first 3D feature grid to a first 3D representation of the first scene. The scene reconstruction model computes a first reconstruction loss based on the first 3D representation and the set of RGBD images. The scene reconstruction model modifies at least one of the first 3D feature grid, a first pre-trained geometry decoder, or a first pre-trained texture decoder based on the first reconstruction loss to generate a second 3D representation of the first scene.

Abstract: The invention relates to a heterophasic propylene copolymer composition comprising a heterophasic propylene copolymer (A), optionally an ethylene-unit containing polymer (B) and an inorganic filler (C), wherein the heterophasic propylene copolymer (A) consists of a) a propylene-based matrix wherein the propylene-based matrix consists of a propylene homopolymer and/or a propylene copolymer consisting of at least 90 wt % of propylene units and at most 10 wt % of comonomer units selected from ethylene monomer units and ?-olefin monomer units having 4 to 10 carbon atoms wherein the propylene-based matrix is present in an amount from 87.0 to 92.0 wt % based on the heterophasic propylene copolymer and b) a dispersed ethylene ?-olefin copolymer, wherein the dispersed ?-olefin copolymer is present in an amount from 8.0 to 13.

Abstract: The present invention discloses a fabrication method and use of a ?40 mm sized fiber optic image inverter, belonging to the field of manufacturing of fiber optic imaging elements. The light-absorbing glass for preparing the ?40 mm sized fiber optic image inverter consists of the following components in molar percentage: SiO2 60-69.9, Al2O3 1.0-10.0, B2O3 10.1-15.0, Na2O 1.0-8.0, K2O 3.0-10.0, MgO 0.1-1.0, CaO 0.5-5.0, ZnO 0-0.1, TiO2 0-0.1, ZrO2 0.1-1.0, Fe2O3 3.0-6.5, Co2O3 0.1-0.5, V2O5 0.51-1.5 and MoO3 0.1-1.0. The ?40 mm sized fiber optic image inverter has the advantages of low crosstalk of stray light, high resolution and high contrast.

Abstract: A method and apparatus for determining gaze information, and an eye-tracking device are provided. The method includes: acquiring two eye images; determining their respective target feature points and their corresponding respective confidence levels; determining first gaze information of a second eye image based on the target feature points of the second eye image when the confidence level corresponding to a first eye image of two eye images is less than a preset confidence level and that the confidence level corresponding to the second eye image is greater than or equal to the preset confidence level, determining the first gaze information as first gaze information of the first eye image; inputting the two eye images respectively into a gaze estimation model to obtain respective second gaze information; and determining target gaze information corresponding to the two eye images based on the first gaze information and the second gaze information.

Abstract: A method and system are provided for tracking instances within a sequence of video frames. The method includes the steps of processing an image frame by a backbone network to generate a set of feature maps, processing the set of feature maps by one or more prediction heads, and analyzing the embedding features corresponding to a set of instances in two or more image frames of the sequence of video frames to establish a one-to-one correlation between instances in different image frames. The one or more prediction heads includes an embedding head configured to generate a set of embedding features corresponding to one or more instances of an object identified in the image frame. The method may also include training the one or more prediction heads using a set of annotated image frames and/or a plurality of sequences of unlabeled video frames.

Abstract: The present application discloses a method for fabricating a fiber optic image inverter with an ultra-short twister. The method includes: drawing a glass rod with a low refractive index and a high strain point temperature into a surrounding pipe fiber; drawing a glass rod with a high refractive index and a high transmittance into a filling glass fiber, and then drawing into a casing pipe absorption fiber; uniformly surrounding an outer side of a cladding glass pipe with the surrounding pipe fiber, and matching a core glass rod and the cladding glass pipe to be drawn into a mono fiber; and then performing fabrication of a multi fiber, fabrication of a multi-multi fiber, heat press fusion and twisting operation in sequence. The fiber optic image inverter with an ultra-short twister with a high resolution and a high contrast and clear imaging is obtained and applied in a low-light-level image intensifier.

Abstract: The present invention provides an ultra-high precision pneumatic force servo system and an intelligent control parameter optimization method therefor.

Abstract: View synthesis is a computer graphics process that generates a new image of a scene from a novel (previously unseen) viewpoint of the scene. Typically, the graphics process relies on a machine learning model that has been trained with ground truth pose information. Since ground truth pose information is not readily available, some solutions rely on a Structure-from-Motion (SfM) library COLMAP to generate pose information for a given image. However, this pre-processing step is not only time-consuming but also can fail due to its sensitivity to feature extraction errors and difficulties in handling texture-less or repetitive regions. The present disclosure provides view synthesis from learned camera poses without relying on SfM pre-processing.

Abstract: The present application discloses a method for fabricating a fiber optic image inverter with an ultra-short twister. The method includes: drawing a glass rod with a low refractive index and a high strain point temperature into a surrounding pipe fiber; drawing a glass rod with a high refractive index and a high transmittance into a filling glass fiber, and then drawing into a casing pipe absorption fiber; uniformly surrounding an outer side of a cladding glass pipe with the surrounding pipe fiber, and matching a core glass rod and the cladding glass pipe to be drawn into a mono fiber; and then performing fabrication of a multi fiber, fabrication of a multi-multi fiber, heat press fusion and twisting operation in sequence. The fiber optic image inverter with an ultra-short twister with a high resolution and a high contrast and clear imaging is obtained and applied in a low-light-level image intensifier.

Abstract: An optical communication device includes a laser, a transmitter (Tx), a receiver (Rx) and a device controller. The laser is configured to generate an optical carrier. The transmitter is configured to generate an optical Tx signal using the optical carrier and to transmit the optical Tx signal to a peer optical communication device. The receiver is configured to receive an optical Rx signal from the peer optical communication device, and to down-convert the optical Rx signal using the optical carrier. The device controller is configured to adjust a frequency of the laser to reduce a Carrier Frequency Offset (CFO) between the received optical Rx signal and the optical carrier generated by the laser, including conditionally applying to a frequency of the laser a series of frequency hops in accordance with a defined dithering sequence.

Abstract: A method for creating a configuration knowledge base, includes: determining a hierarchical structure between a plurality of modules in a product family based on the plurality of modules; determining a knowledge package of each module based on the hierarchical structure; and releasing the knowledge package of each module in the product family to a configuration knowledge base, to obtain a configuration knowledge base of the product family. The hierarchical structure indicates a hierarchical relationship between a parent module and a child module in the plurality of modules. Each parent module includes one or more child modules. A knowledge package of the parent module includes an input interface algorithm of the child module associated with the parent module.

Abstract: A cyclical and continuous device and method for wet sieving of polysaccharide granules are provided. The device includes a first dosing tank, a second dosing tank connected to the first dosing tank by a first circulation pump, a first suction filtration tank connected to the second dosing tank by a second circulation pump, and a second suction filtration tank connected to the first dosing tank by a third circulation pump. An overflow port provided on a side wall of the second dosing tank is higher than a feed port of the first dosing tank. A sieve of a mesh filter for coarse powder is disposed in the first dosing tank. A sieve of a mesh filter for fine powder is disposed in the second dosing tank. A stirring device is disposed in both the first dosing tank and the second dosing tank.

Abstract: A method and apparatus for controlling an image acquisition device are provided. The method includes that: multiple images photographed by multiple image acquisition devices are acquired; an active image acquisition device and an inactive image acquisition device in the multiple image acquisition devices are determined according to the multiple images; the active image acquisition device is controlled to photograph a first image by using a first configuration, and the inactive image acquisition device is controlled to photograph a second image by using a second configuration; the first image transmitted by the active image acquisition device is acquired, and the second image transmitted by the inactive image acquisition device is acquired, and a bandwidth required by the active image acquisition device to transmit the first image is greater than a bandwidth required by the inactive image acquisition device to transmit the second image.

Abstract: The present invention provides a method for removing an organic solvent from starch hemostatic microspheres, comprising the following steps: 1. taking to-be-dried starch hemostatic microspheres and laying them flatly on drying trays with attention to laying them as uniformly and thinly as possible; 2. taking an adsorbent and subpackaging it into dialyzing paper bags for sealing; and 3. placing the trays and the dialyzing paper bags completed in the previous two steps on separators of a low-temperature vacuum oven in layers, setting the oven temperature at 0-20° C., then vacuumizing and keeping pressure for 15-48 hours. The method provided by the present invention can reduce organic solvent residue in the starch hemostatic microspheres to less than 0.05%, which meets the requirements of relevant standards for medical devices, thereby improving safety of products.

Abstract: A coating composition for reducing heat gain of an article includes TiO2 particles; hollow glass microspheres; a phosphor material having a quantum yield of at least about 90% and an excitation wavelength from about 250 nm to about 440 nm, matching those of the TiO2 particles; and a polymer. Also a method for reducing heat gain of an article by using the coating composition of the present invention.

Abstract: In various embodiments, a scene reconstruction model generates three-dimensional (3D) representations of scenes. The scene reconstruction model computes a first 3D feature grid based on a set of red, blue, green, and depth (RGBD) images associated with a first scene. The scene reconstruction model maps the first 3D feature grid to a first 3D representation of the first scene. The scene reconstruction model computes a first reconstruction loss based on the first 3D representation and the set of RGBD images. The scene reconstruction model modifies at least one of the first 3D feature grid, a first pre-trained geometry decoder, or a first pre-trained texture decoder based on the first reconstruction loss to generate a second 3D representation of the first scene.

Abstract: In various embodiments, a scene reconstruction model generates three-dimensional (3D) representations of scenes. The scene reconstruction model maps a first red, blue, green, and depth (RGBD) image associated with both a first scene and a first viewpoint to a first surface representation of at least a first portion of the first scene. The scene reconstruction model maps a second RGBD image associated with both the first scene and a second viewpoint to a second surface representation of at least a second portion of the first scene. The scene reconstruction model aggregates at least the first surface representation and the second surface representation in a 3D space to generate a first fused surface representation of the first scene. The scene reconstruction model maps the first fused surface representation of the first scene to a 3D representation of the first scene.

Abstract: In various embodiments, a training application trains a machine learning model to generate three-dimensional (3D) representations of two-dimensional images. The training application maps a depth image and a viewpoint to signed distance function (SDF) values associated with 3D query points. The training application maps a red, blue, and green (RGB) image to radiance values associated with the 3DI query points. The training application computes a red, blue, green, and depth (RGBD) reconstruction loss based on at least the SDF values and the radiance values. The training application modifies at least one of a pre-trained geometry encoder, a pre-trained geometry decoder, an untrained texture encoder, or an untrained texture decoder based on the RGBD reconstruction loss to generate a trained machine learning model that generates 3D representations of RGBD images.