FACIAL PARSING METHOD AND RELATED DEVICES
    4.
    发明公开

    公开(公告)号:EP4047509A1

    公开(公告)日:2022-08-24

    申请号:EP20889944.3

    申请日:2020-08-18

    IPC分类号: G06K9/00

    摘要: A facial parsing method and apparatus, a facial parsing network training method and apparatus, an electronic device and a non-transitory computer-readable storage medium, which relate to the field of artificial intelligence. The facial parsing method comprises: inputting a facial image into a pre-trained facial parsing neural network (S201); extracting semantic features of the facial image by using a semantic perception sub-network (S202); extracting boundary features of the facial image by using a boundary perception sub-network (S203); and processing cascaded semantic features and boundary features by using a fusion sub-network, so as to obtain a facial region to which each pixel in the facial image belongs (S204). The method can improve the resolution capability of a neural network for boundary pixels between different facial regions of a facial image, thereby improving the precision of facial parsing.

    APPARATUS AND METHOD FOR DISTRIBUTION ARTICLE PLACEMENT DETECTION, ROBOT, DISTRIBUTION DEVICE, AND CONTROLLER

    公开(公告)号:EP4005967A1

    公开(公告)日:2022-06-01

    申请号:EP20864806.3

    申请日:2020-09-11

    发明人: XU, Zhetao XIAO, Jun

    IPC分类号: B66F9/07

    摘要: An apparatus and a method for distribution object placement detection, a robot, a distribution apparatus, and a controller, relating to the field of logistics. Said apparatus comprises : a plurality of distance measurement sensors, arranged on an end surface of a placement area facing a distribution object on a protruding part (21) of a robot. The protruding part (21) is positioned on a chassis (22) of the robot, and each distance measurement sensor among the plurality of distance measurement sensors is configured to measure the distance between each distance measurement sensor and an end surface facing the plurality of distance measurement sensors of a distribution object. The controller (13) is configured to determine, on the basis of a condition that the distance measured by each distance measurement sensor is no greater than a first distance threshold, that the distribution object has been positioned at a preset position of the chassis (22). The present apparatus for distribution object placement detection is able to improve precision in detecting whether a distribution object is positioned at a preset position of a robot, thus reducing loss in a shipping process.

    DOME SETTING FORCE ADJUSTING DEVICE FOR A CAN BOTTOM FORMER

    公开(公告)号:EP3895825A1

    公开(公告)日:2021-10-20

    申请号:EP21169220.7

    申请日:2018-01-12

    发明人: SWEDBERG, Rick

    IPC分类号: B21D51/26 B21D22/30 B21D22/28

    摘要: A device for adjusting a dome setting force of a can bottom former comprising a movable housing, the device comprising:
    an outer end plate;
    an inner end plate movably mounted proximate the outer end plate;
    a dome-setting spring positioned between the inner end plate and the movable housing; and
    an adjustment means for exerting a displacing force on the inner end plate toward the movable housing.

    WEBPAGE ILLUSTRATION PROCESSING METHOD, SYSTEM AND DEVICE, AND STORAGE MEDIUM

    公开(公告)号:EP3872655A1

    公开(公告)日:2021-09-01

    申请号:EP20769750.9

    申请日:2020-01-02

    IPC分类号: G06F17/00

    摘要: There discloses a webpage illustration processing method, system and device, and a storage medium. The processing method comprises: acquiring a first image set corresponding to a product category (S 101); labelling at least one region of interest for each image in the first image set, wherein each region of interest is used for representing an object (S 102); acquiring a scanning window (S 103); using the scanning window to scan the regions of interest, acquiring scanning results, and placing the scanning results into a sample training set (S 104); establishing an illustration frame acquisition model by taking the first image set as input and taking the sample training set as output (S 105); acquiring an image to be processed, and using the illustration frame acquisition model to acquire a target webpage illustration frame corresponding to the image to be processed (S 106); and cutting, according to the target webpage illustration frame, the image to be processed so as to obtain a target webpage illustration (S 107). By means of a deep learning algorithm for image detection and classification, a target illustration suitable for serving as a webpage illustration is directly cut from an original natural scene image, such that manual labelling can be reduced while the availability of a target illustration is ensured, thereby reducing time costs and improving working efficiency.