Beijing Sensetime Technology Development Co., Ltd.

Huawei Technologies Co., Limited

Chinese Academy of Sciences

Wanli Ouyang

Xiaogang Wang

Xiao Chu

Jianzhuang Liu

Hongsheng Li

Dong Xu

Huchuan Lu

Lijun Wang

Wei Li

Wei Yang

Yu Liu

Jie Cao

Zhenan Sun

Hongyang Li

Wenqi Ju

Xiaoyun Zhang

Chunlei Cai

Hongwen Zhang

Zhiyong Gao

Rui Su

Guo Lu

Luping Zhou

Disclosed are methods, apparatuses and systems for CNN network adaption and object online tracking. The CNN network adaption method comprises: transforming a first feature map into a plurality of sub-feature maps, wherein the first feature map is generated by the pre-trained CNN according to a frame of the target video; convolving each of the sub-feature maps with one of a plurality of adaptive convolution kernels, respectively, to output a plurality of second feature maps with improved adaptability; training, frame by frame, the adaptive convolution kernels.

Methods and systems for CNN network adaption and object online tracking

A reconstruction method of a three-dimensional (3D) human body model includes: acquiring, by a fully convolutional network (FCN) module, a global UVI map and a local UVI map of a body part according to a human body image (S); estimating, by a first neural network, a camera parameter and a shape parameter of the human body model based on the global UVI map (S); extracting, by a second neural network, rotation features of joints of a human body based on the local UVI map (S); refining, by using a position-aided feature refinement strategy, the rotation features of the joints of the human body to acquire refined rotation features (S); and estimating, by a third neural network, a pose parameter of the human body model based on the refined rotation features (S). The reconstruction method achieves accurate and efficient reconstruction of the human body model, and improves robustness of pose estimation.

Reconstruction method of three-dimensional (3D) human body model, storage device and control device

An object detection method, a neural network training method, an apparatus, and an electronic device include: obtaining, through prediction, multiple fused feature graphs from images to be processed, through a deep convolution neural network for target region frame detection, obtaining multiple first feature graphs from a first subnet having at least one lower sampling layer, obtaining multiple second feature graphs from a second subnet having at least one upper sampling layer, and obtaining fused graph by fusing multiple first feature graphs and multiple second feature graphs respectively; and obtaining target region frame data according to the multiple fused feature graphs. Because the fused feature graphs better represent semantic features on high levels and detail features on low levels in images, target region frame data of big and small objects in images can be effectively extracted according to the fused feature graphs, thereby improving accuracy and robustness of object detection.

Method and apparatus for detecting object, method and apparatus for training neural network, and electronic device

A method for video compression processing, an electronic device, and a storage medium. In the method for video compression processing, video compression is performed by using an end-to-end model trained with a single loss function, to obtain video compression information.

Method for video compression processing, electronic device and storage medium

A target detection method and apparatus, a target detection device and a storage medium. The target detection method comprises: acquiring n types of detection frames which are determined by n different target detectors and which are related to a positioning target, wherein the n different target detectors may extract n different types of image features from the same image, and n is a positive integer equal to or greater than 2; obtaining an integrated detection frame on the basis of the n types of detection frames; each target detector extracting image features from an input image on the basis of the integrated detection frame to obtain n types of image features; performing image feature optimization on the basis of the n types of image features to obtain optimized image features; and obtaining an optimized detection frame according to the optimized image features so as to detect a target.

Target detection method and apparatus, device, and storage medium

A method and a system for pose estimation are provided. The method includes: extracting a plurality of sets of part-feature maps from an image, each set of the extracted part-feature maps encoding the messages for a particular body part and forming a node of a part-feature network; passing a message of each set of the extracted part-feature maps through the part-feature network to update the extracted part-feature maps, resulting in each set of the extracted part-feature maps incorporating the message of upstream nodes; estimating, based on the updated part-feature maps, the body part within the image.

Method and system for pose estimation

Disclosed are methods for object tracking. In an example, the method comprises: determining a region of interest (ROI) in a first frame of a video sequences; feeding the determined ROI forward through a first CNN (convolutional network) to obtain a plurality of first feature maps in a higher layer of the CNN and a plurality of second feature maps in a lower layer of the first CNN; selecting a plurality of feature maps from the first and second feature maps, respectively; predicting, based on the selected first and second feature maps, two target heat maps indicating a target location for said objects in the current frame, respectively; and estimating, based on the two predicated target heat maps, a final target location for the object in the current frame.

Systems and methods for object tracking

The present application relates to post estimation methods and apparatuses, and computer systems, wherein the method includes: extracting a feature in each body part of body parts of an object in an image to generate feature maps for the each body part, wherein the each body part corresponds to at least one part type; predicting part type scores respectively for features in each of the feature maps and establishing part type score maps based on the predicted part type scores; optimizing, for at least one point in each of the part type score maps, the established part type score maps based on a message related to an adjacent body part of a body part corresponding to the point; and determining an estimated position and an estimated part type of the each body part based on the optimized part type score maps to obtain an estimated pose of the object.

Posture estimation method and apparatus, and computer system

A body relationship estimation method and apparatus are disclosed. The method includes obtaining a target picture, calculating a first body relationship feature of two persons according to at least one of first location information of a body part of each person of the two persons in the target picture or second location information of body parts of the two persons, where the first location information is obtained by performing single-person gesture estimation on each person, and the second location information is obtained by performing two-person joint gesture estimation on the two persons when the first location information indicates that the body parts of the two persons overlap, and determining a body relationship between the two persons according to the first body relationship feature.

Body relationship estimation method and apparatus

An image characteristic estimation method and device is presented, where content of the method includes extracting at least two eigenvalues of input image data, and executing the following operations for each extracted eigenvalue, until execution for the extracted eigenvalues is completed. Selecting an eigenvalue, and performing at least two matrix transformations on the eigenvalue using a pre-obtained matrix parameter in order to obtain a first matrix vector corresponding to the eigenvalue; when a first matrix vector corresponding to each extracted eigenvalue is obtained, obtaining second matrix vectors with respect to the at least two extracted eigenvalues using a convolutional network calculation method according to the obtained first matrix vector corresponding to each eigenvalue; and obtaining a status of an image characteristic in the image data by means of estimation according to the second matrix vectors. In this way, accuracy of estimation is effectively improved.

Growing community of inventors

Beijing, China

Wanli Ouyang