Huawei Technologies Co., Limited

Yang Wang

Jin Tang

Yuanhao Yu

Wentao Liu

Zhixiang Chi

Rui Xiang Chai

Irina Kezele

Zheng Liu

Juwei Lu

Roel Vertegaal

Rasoul Mohammadi Nasiri

Seyed Shahabeddin Nabavi

Hisham Elhalabi

Zhenhao Li

Li Gu

Seyed Mehdi Ayyoubzadeh

Huan Liu

Seel Nimeshkumar Patel

Xiaolin Wu

Amartya Mukherjee

Sheral Sweta Kumar

Mostafa Shahabinejad

There is provided a method, apparatus and system for adapting a machine learning model for optical flow prediction. A machine learning model can be trained or adapted based on compressed video data, using motion vector information extracted from the compressed video data as ground-truth information for use in adapting the model to a motion vector prediction task. The model so adapted can accordingly be adapted for the similar task of optical flow prediction. Thus, the model can be adapted at test time to image data which is taken from an appropriate distribution. A meta-learning process can be performed prior to such model adaptation to potentially improve the model's performance.

Method, apparatus and system for adaptating a machine learning model for optical flow map prediction

Methods, devices and computer-readable media for processing a compressed video to perform an inference task are disclosed. Processing the compressed video may include selecting a subset of frame encodings of the compressed video, or zero or more modalities (RGB, motion vectors, residuals) of a frame encoding, for further processing to perform the inference task. Pre-existing motion vector and/or residual information in frame encodings of the compressed video are leveraged to adaptively and efficiently perform the inference task. In some embodiments, the inference task is an action recognition task, such as a human action recognition task.

Method, device, and medium for adaptive inference in compressed video domain

Systems, methods and computer-readable medium for predicting a depth for a video frame are disclosed. An example method may include steps of: receiving a plurality of training data, each comprising a set of consecutive video frames and a depth representation of a subsequent video frame to the consecutive video frames; receiving a pre-trained neural network model fhaving a plurality of weights θ; while the pre-trained neural network model fhas not converged: computing a plurality of second weights θ', based on each set of consecutive video frames, and updating the plurality of weights θ, based on the plurality of training data and the plurality of second weights θ′; receiving a plurality of new consecutive video frames with consecutive timestamps; and predicting a depth representation of video frame immediately subsequent to the new consecutive video frames based on the updated plurality of weights θ.

Methods, systems and computer medium for scene-adaptive future depth prediction in monocular videos

This disclosure provides for methods and system for meta few-shot class incremental learning. According to an aspect a method is provided. The method includes obtaining at least one weight attention map of a first network and updating weights of a second network using the at least one weight attention map, where the second network is a modulatory network. The method further includes generating at least one feature attention map of the second network based on the at least one weight attention map of the first network and a set of input images of at least one class. The method further includes generating at least one feature map of the first network based on the set of input images of the at least one class, and updating the at least one feature map of the first network based on the feature attention map of the second network.

Meta few-shot class incremental learning

A method, device and computer-readable medium for generating a super-resolution version of a compressed video stream. By leveraging the motion information and residual information in compressed video streams, described examples are able to skip the time-consuming motion-estimation step for most frames and make the most use of the SR results of key frames. A key frame SR module generates SR versions of I-frames and other key frames of a compressed video stream using techniques similar to existing multi-frame approaches to VSR. A non-key frame SR module generates SR version of the non-key inter frames between these key frames by making use of motion information and residual information used to encode the inter frames in the compressed video stream.

Method, device, and medium for generating super-resolution video

Methods and systems for image deblurring are described. The weights of a deblurring network are first trained by meta-training the deblurring network on both a primary deblurring task and an auxiliary reconstruction task. Application-time training of the deblurring network is then performed, using an application-time blurry input image, to obtain values of application-time trained weights. Application-time training includes performing the auxiliary reconstruction task on the application-time blurry input image, and updating the weights of the deblurring network based on an auxiliary loss computed from the auxiliary reconstruction task. A deblurred output image is generated from the application-time blurry input image, using the application-time trained weights in the deblurring network.

Methods and systems for deblurring blurry images

Devices and methods for extracting motion vector data during decoding compressed of video data are described. At a video decoder, an encoded video data for a frame of video from an input buffer is obtained. The encoded video data is decoded to obtain decoded image data for a decoded frame, where the decoding includes extracting corresponding motion vector data for the decoded frame. The decoded image data is stored in a temporary storage indexed with a given index, and the corresponding motion vector data is stored in a same or different temporary storage indexed with the given index. An output buffer indexed with the given index is filled with the decoded image data and the corresponding motion vector data stored in the respective temporary storage indexed with the given index.

Methods and devices for extracting motion vector data from compressed video data

A system and a method for performing 2D on-screen user gaze estimation using an input facial image of a user captured using a camera associated to a processing device having a display. The method and system allow automated user calibration through automatic recording of calibration samples each including a calibration facial image of the user and an interaction point corresponding to a point on the display where an occurrence of a user interaction was detected when the corresponding calibration image was captured. The system and method also optimize user-specific parameters using the calibration samples by iteratively minimizing a total difference between the interaction points of a plurality of the calibration samples and corresponding 2D gaze estimation results and convert an estimated 3D gaze direction into a 2D gaze estimation result corresponding to a point on the display, by applying the users-specific parameters.

Growing community of inventors

Winnipeg, Canada

Yang Wang