31results about How to "Troubleshoot technical issues that aren't working" patented technology

The present application discloses a target tracking method and device. The method includes extracting image features in an image target candidate area of ​​a video frame, and outputting an initial response value; smoothing the initial response value of the video frame; and updating the response value to determine the position information of the target in the video frame. The present application solves the technical problem of poor effect of the target tracking method. Through the method of this application, the output response of the network will often appear multi-peak and drift due to complex target states such as scale transformation and sudden movement. By fully considering the similarity relationship between adjacent superpixel blocks of the image, The relationship is used to constrain the initial response value, suppress the drift phenomenon in the tracking process, and then correct the position information of the target. Furthermore, the tracking accuracy and tracking efficiency of the object tracking method are improved.

The invention proposes an interactive service recommendation method with dual interaction relationship and attention mechanism, the core of which is a service call prediction model. The model includes: using the Google Inception network-based text feature extractor and the graph embedding technology node2vec to extract the content features and call features of Web applications and services; based on the attention mechanism and multi-layer perceptron, respectively learn the Web applications to be developed , the content interaction and call interaction between the selected service and the candidate service; the above two interactions are integrated based on the multi-layer perceptron; the softmax function is used to predict the possibility that the developer selects the current candidate service as the component service. The interactive service recommendation method proposed by the present invention includes: using the above model to predict the score of each candidate service; recommending the N candidate services with the highest predicted scores to the developer; after the developer selects several component services, the recommendation process enters the next step round; repeat the above steps until the developer's requirements are fully realized.