Dell Products Inc.

Emc IP Holding Company LLC

Zijia Wang

Zhen Jia

Jiacheng Ni

Zhisong Liu

Jinpeng Liu

Qiang Chen

Wenbin Yang

Min Gong

Bin He

Danqing Sha

Sanping Li

Kenneth Durazzo

Zhenzhen Lin

Chenxi Hu

Tianxiang Chen

GuangZhou Zhou

Pengfei Wu

Victor Fong

Si Ruo Chen

Amy N Seibel

Eric Joseph Bruno

Anzhou Hou

Xuwei Tang

Jin Ru Yan

Chun Xi Chen

Wenlei Wu

Bo Wei

ChunXi Chen

Yufeng Wang

Si Chen

Methods, devices and computer program products for processing video are disclosed herein. A method includes: generating, based on a reference image and a first frame of a video comprising an object, a two-dimensional avatar image of the object; and generating a base three-dimensional avatar of the object by performing a three-dimensional transformation on the two-dimensional avatar image and the object in the first frame. The method further includes: generating a three-dimensional avatar video corresponding to the video based on the base three-dimensional avatar and features of the video, the features comprising differences of the object between adjacent frames of the video. This solution enables the generation of a customized three-dimensional avatar video for an object in a video, where the avatar can move in synchronization with the object and retain the unique features of the object, and can provide a more detailed and vivid representation than a two-dimensional avatar.

Method, device, and computer program product for processing video

Embodiments of the present disclosure relate to a method, an electronic device, and a computer program product for acquiring an image. The method includes distilling an original image set through a capsule neural network model to generate a distilled image set, wherein the distilled image set includes a plurality of distilled images. The method further includes acquiring a first feature of a first image through the capsule neural network model. The method further includes acquiring a plurality of distilling features of the plurality of distilled images respectively through the capsule neural network model. The method further includes determining a plurality of similarities between the first feature and the plurality of distilling features respectively. The method further includes acquiring at least one original image matching the first image based on the plurality of similarities.

Method, electronic device, and computer program product for acquiring image

Embodiments of the present disclosure relate to a method, a device, and a computer program product for text to speech. The method includes encoding a reference waveform of a first speaker to obtain an encoded style feature separated from a second speaker. The method further includes transferring the encoded style feature to a spectrogram obtained by encoding an input text, to obtain a style transferred spectrogram. The method further includes converting the style transferred spectrogram into a time-domain speech waveform. According to the method for text to speech in the present disclosure, a comparative learning framework can also flexibly and effectively synthesize speech with a style of a target speaker, thus realizing lightweight speech style transfer, making it possible to learn high-quality and recognizable features of speech synthesis, and realizing effective speaker feature learning. In addition, the model will be beneficial to other downstream tasks.

Method, device, and computer program product for text to speech

Embodiments disclosed herein relate to a method, an electronic device, and a computer program product for data anonymization. The method includes: performing classification on data by a classifier to obtain data types of the data. The method further includes: performing anonymization on the data by a first anonymization model to obtain first anonymized data. The method further includes: determining, based on the data types, using an anonymizer whether re-anonymization needs to be performed on the first anonymized data. The method further includes: performing, based on a determination that the re-anonymization needs to be performed, the re-anonymization on the first anonymized data by a second anonymization model to obtain second anonymized data. Accordingly, anonymization processing may be performed on data using different anonymization models for different types of data to obtain the final anonymized data and to ensure that no data leakage occurs.

Method, electronic device, and computer program product for data anonymization

The present disclosure relates to a method, a device, and a computer program product for model comparison. The method includes generating a detection image based on an original image. The method further includes obtaining a first classification result by sending the detection image to a target model, and obtaining a second classification result by sending the detection image to a to-be-detected model. In addition, the method further includes comparing the first classification result with the second classification result, and determining, in response to the first classification result being the same as the second classification result, that the target model is the same as the to-be-detected model. The method of the present disclosure can verify whether the to-be-detected model plagiarizes the target model without knowing any internal structure, parameters, weights, and other information of the to-be-detected model.

Method, device, and computer program product for model comparison

Embodiments of the present disclosure relate to a method, an electronic device, and a computer program product for generating a three-dimensional (3D) model. The method includes generating two-dimensional (2D) features of a 2D image on the basis of performing feature extraction on the 2D image. The method further includes generating, on the basis of the 2D features by using a symmetric conversion module, 2D features subjected to symmetric conversion, wherein the symmetric conversion module is configured to generate 2D features of an object in the 2D image in different viewing angles. The method further includes generating a 3D model of the 2D image on the basis of the 2D features subjected to symmetric conversion. By means of embodiments of the present disclosure, the generation of the 3D model can be supervised by using the symmetric conversion module, thus achieving generation of a 3D model with a better effect.

Method, electronic device, and computer program product for generating three-dimensional model

Embodiments of the present disclosure relate to a question answering method, an electronic device, and a computer program product. The method includes: determining an answer library associated with a question; determining a restricted Boltzmann machine associated with the answer library, wherein the restricted Boltzmann machine is configured to determine a set of questions that the answer library can answer and association relationships between questions in the set of questions and the answer library; and determining, using the restricted Boltzmann machine, annotation information associated with the question and targeted to the answer library. With the technical solution of the present disclosure, it is possible to determine, while determining an answer library associated with the question, annotation information that is targeted to the answer library, and to enable customer service personnel to have a more thorough understanding of the determined answer library and obtain targeted recommendation information.

Question answering method, electronic device, and program product

Embodiments of the present disclosure provide a method, an electronic device, and a computer program product for analyzing samples. The method includes acquiring a set of feature representations associated with a set of samples. The set of samples illustratively have classification information for indicating classifications of the set of samples. The method further includes adjusting the set of feature representations so that distances between feature representations of samples corresponding to the same classification are less than a first distance threshold. The method further includes training a classification model based on the adjusted set of feature representations and the classification information. The classification model is illustratively configured to receive an input sample and determine a classification of the input sample. In this manner, a relatively accurate classification model can be trained using a small number of samples, thereby reducing computation time and required computation capacity.

Method, electronic device, and computer program product for analyzing samples

Embodiments of the present disclosure relate to a method, an electronic device, and a computer program product for verifying a virtual avatar. The method includes: determining, based on the virtual avatar and a plurality of reference images in an image library, a first similarity set comprising a plurality of similarities between the virtual avatar and the plurality of reference images; and selecting a plurality of candidate images from the plurality of reference images based on the first similarity set and a predetermined number. The method further includes: determining a second similarity set comprising a plurality of similarities between the virtual avatar and the plurality of candidate images; and determining a verification result for the virtual avatar based on a subset in the first similarity set that is associated with the plurality of candidate images, and the second similarity set.

Method, electronic device, and computer program product for verifying virtual avatar

A method includes generating a feature map of a second resolution for an input image of a first resolution, the first resolution being less than the second resolution. The method further includes generating a feature map of the second resolution and a feature map of the first resolution through a first residual-based dense back projection (RDBP) network. In addition, the method further includes generating a feature map of the second resolution and a feature map of the first resolution through a second RDBP network, the first RDBP network and the second RDBP network respectively including a downsampling back projection layer, an upsampling back projection layer, and an attention layer. The method further includes generating an output image of the second resolution based on the feature maps of the second resolution and the feature maps of the first resolution generated through the first RDBP network and the second RDBP network.

Method, device, and computer program product for image processing

Embodiments of the present disclosure involve a method, an electronic device, and a computer program product for processing data. The method includes detecting a type of the data, wherein the data includes sensitive data. The method further includes selecting, according to the detected type, a generative model corresponding to the detected type from a plurality of generative models. The method further includes processing the data by using the selected generative model to desensitize the sensitive data from the data. The method for processing data according to embodiments of the present disclosure processes the data by combining various types of generative models, and the application potentials of various generative models can be fully exploited, so that desensitization processing of a plurality of types of data can be realized more quickly and accurately, and leakage of sensitive information can be avoided, thereby ensuring the security of various types of data.

Method, electronic device, and computer program product for processing data

Embodiments of the present disclosure relate to a model optimization method, an electronic device, and a computer program product. This method includes: determining an initial learning rate combination for a deep learning model, wherein the initial learning rate combination includes a plurality of learning rates, each learning rate being determined for one of a plurality of layers of the deep learning model, and the plurality of learning rates including static learning rates and dynamic learning rates; and adjusting the initial learning rate combination to obtain a target learning rate combination, wherein an accuracy rate achieved when the target learning rate combination is used to train the deep learning model is higher than or equal to a first threshold accuracy rate. With the technical solution of the present disclosure, a deep learning model can be optimized by setting learning rates for each layer of the deep learning model.

Model optimization method, electronic device, and computer program product

Embodiments of the present disclosure provide a method, an electronic device, and a computer program product for switching a chat object. The method includes: using a robot as a chat object to converse with a user; outputting a target sentiment score based on current discourse, historical discourse, and predicted discourse associated with the user; and switching the chat object conversing with the user from the robot to human in response to the target sentiment score indicating that the user's sentiment is a negative sentiment. Implementation of this method can improve human-machine interactivity and communication efficiency, and ensure the sustainability of communication between a chat system and a user, thus improving the user's satisfaction with a service system.

Method, electronic device, and computer program product for switching chat object

Embodiments of the present disclosure relate to a method, a device, and a computer program product for image processing. The method includes: obtaining a target object segmentation map for a target image and a reference dress-up segmentation map for a reference image, wherein the target image presents a target object, and the reference image presents reference dress-up; extracting a target image feature of the target image and a reference image feature of the reference image by means of image encoding of the target image and the reference image; and transferring the reference dress-up in the reference image to the target object based on the target object segmentation map, the reference dress-up segmentation map, the target image feature, and the reference image feature.

Embodiments of the present disclosure relate to a method, a device, and a computer program product for rendering. The method includes obtaining features of a three-dimensional (3D) graph based on the 3D graph and a pose of a camera capturing the 3D graph. The method further includes generating a plurality of two-dimensional (2D) subgraphs for the 3D graph based on the features of the 3D graph, wherein the plurality of 2D subgraphs are adjustable. The method further includes rendering a scene in the 3D graph based on the generated plurality of 2D subgraphs. In this way, customization of a rendering process can be achieved, and visual quality during scene decomposition and synthesis can be improved, thereby enhancing 3D reconstruction.

Growing community of inventors

Shanghai, China

Zijia Wang