Qualcomm Incorporated

Qualcomm Technologies, Inc.

Taco Sebastiaan Cohen

Yang Yang

Reza Pourreza

Amir Said

Yinhao Zhu

Amirhossein Habibian

Ties Jehan Van Rozendaal

Max Welling

Reza Pourreza Shahri

Yizhe Zhang

Hoang Cong Minh Le

Yadong Lu

Pim De Haan

Auke Joris Wiggers

Markus Nagel

Berkay Kicanaoglu

Dana Kianfar

Jongha Ryu

Gabriele Cesa

Guillaume Konrad Sautière

Yunfan Zhang

Johann Hinrich Brehmer

Guillaume Konrad Sautiere

Maurice Weiler

Adam Waldemar Golinski

Mirgahney Husham Awadelkareem Mohamed

Iris Anne Marie Huijben

Certain aspects of the present disclosure provide a method of performing machine learning, comprising: generating a neural network model; and training the neural network model for a task with a first set of input data, wherein: the training uses a total loss functionincluding an equivariance loss componentaccording to=+α, and α>0.

Data and compute efficient equivariant convolutional networks

Techniques are described for compressing and decompressing data using machine learning systems. An example process can include receiving a plurality of images for compression by a neural network compression system. The process can include determining, based on a first image from the plurality of images, a first plurality of weight values associated with a first model of the neural network compression system. The process can include generating a first bitstream comprising a compressed version of the first plurality of weight values. The process can include outputting the first bitstream for transmission to a receiver.

Implicit image and video compression using machine learning systems

Techniques are described for processing video data, such as by performing learned bidirectional coding using a unidirectional coding system and an interpolated reference frame. For example, a process can include obtaining a first reference frame and a second reference frame. The process can include generating a third reference frame at least in part by performing interpolation between the first reference frame and the second reference frame. The process can include performing unidirectional inter-prediction on an input frame based on the third reference frame, such as by estimating motion between an input frame and the third reference frame, and generating a warped frame at least in part by warping one or more pixels of the third reference frame based on the estimated motion. The process can include generating, based on the warped frame and a predicted residual, a reconstructed frame representing the input frame, the reconstructed frame including a bidirectionally-predicted frame.

Learned B-frame coding using P-frame coding system

Certain aspects of the present disclosure are directed to methods and apparatus for compressing video content using deep generative models. One example method generally includes receiving video content for compression. The received video content is generally encoded into a latent code space through an encoder, which may be implemented by a first artificial neural network. A compressed version of the encoded video content is generally generated through a trained probabilistic model, which may be implemented by a second artificial neural network, and output for transmission.

Video compression using deep generative models

Certain aspects of the present disclosure provide a method for performing machine learning, comprising: determining a plurality of vertices in a neighborhood associated with a mesh including a target vertex; determining a linear transformation configured to parallel transport signals along all edges in the mesh to the target vertex; applying the linear transformation to the plurality of vertices in the neighborhood to form a combined signal at the target vertex; determining a set of basis filters; linearly combining the basis filters using a set of learned parameters to form a gauge equivariant convolution filter, wherein the gauge equivariant convolution filter is constrained to maintain gauge equivariance; applying the gauge equivariant convolution filter to the combined signal to form an intermediate output; and applying a nonlinearity to the intermediate output to form a convolution output.

Gauge equivariant geometric graph convolutional neural network

A processor-implemented method for image compression using an artificial neural network (ANN) includes receiving, at an encoder of the ANN, an image and a spatial segmentation map corresponding to the image. The spatial segmentation map indicates one or more regions of interest. The encoder compresses the image according to a controllable spatial bit allocation. The controllable spatial bit allocation is based on a learned quantization bin size.

Neural image compression with controllable spatial bit allocation

Systems and techniques are described herein for processing media data using a neural network system. For instance, a process can include obtaining a latent representation of a frame of encoded image data and generating, by a plurality of decoder transformer layers of a decoder sub-network using the latent representation of the frame of encoded image data as input, a frame of decoded image data. At least one decoder transformer layer of the plurality of decoder transformer layers includes: one or more transformer blocks for generating one or more patches of features and determine self-attention locally within one or more window partitions and shifted window partitions applied over the one or more patches; and a patch un-merging engine for decreasing a respective size of each patch of the one or more patches.

Transformer-based architecture for transform coding of media

A method for generating a convolutional neural network to operate on a spherical manifold, generates locally-defined gauges at multiple positions on the spherical manifold. A convolution is defined at each of the positions on the spherical manifold with respect to an arbitrarily selected locally-defined gauge. The results of the convolution that is defined at each position based on gauge equivariance is translated to obtain a manifold convolution.

Icospherical gauge convolutional neural network

A video encoder determines scaled transform coefficients, wherein determining the scaled transform coefficients comprises scaling transform coefficients of a block of the video data according to a given quantization step. The video encoder determines scalar quantized coefficients, wherein determining the scalar quantized coefficients comprises applying scalar quantization to the scaled transform coefficients of the block. Additionally, the video encoder applies a neural network that determines a respective set of probabilities for each respective transform coefficient of the block. The respective set of probabilities for the respective transform coefficient includes a respective probability value for each possible adjustment value in a plurality of possible adjustment values. Inputs to the neural network include the scaled transform coefficients and the scalar quantized coefficients. The video encoder determines, based on the set of probabilities for a particular transform coefficient of the block, a quantization level for the particular transform coefficient.

Parallelized rate-distortion optimized quantization using deep learning

Certain aspects of the present disclosure provide techniques for compressing content using a neural network. An example method generally includes receiving content for compression. The content is encoded into a first latent code space through an encoder implemented by an artificial neural network trained to generate a latent space representation of the content. A first compressed version of the encoded content is generated using a first quantization bin size of a series of quantization bin sizes. A refined compressed version of the encoded content is generated by scaling the first compressed version of the encoded content into one or more second quantization bin sizes smaller than the first quantization bin size, conditioned at least on a value of the first compressed version of the encoded content. The refined compressed version of the encoded content is output for transmission.

Progressive data compression using artificial neural networks

Certain aspects of the present disclosure are directed to methods and apparatus for compressing video content using deep generative models. One example method generally includes receiving video content for compression. The received video content is generally encoded into a latent code space through an auto-encoder, which may be implemented by a first artificial neural network. A compressed version of the encoded video content is generally generated through a trained probabilistic model, which may be implemented by a second artificial neural network, and output for transmission.

A computer-implemented method for operating an artificial neural network (ANN) includes receiving an input by the ANN. The ANN generates a latent representation of the input. The latent representation is communicated according to a bit rate based on a learned latent scaling parameter. The latent scaling parameter is learned based on a channel index and a tradeoff parameter value that corresponds to a value that balances the bit rate and a distortion.

Variable bit rate compression using neural network models

Techniques are described for compressing data using machine learning systems and tuning machine learning systems for compressing the data. An example process can include receiving, by a neural network compression system (e.g., trained on a training dataset), input data for compression by the neural network compression system. The process can include determining a set of updates for the neural network compression system, the set of updates including updated model parameters tuned using the input data. The process can include generating, by the neural network compression system using a latent prior, a first bitstream including a compressed version of the input data. The process can further include generating, by the neural network compression system using the latent prior and a model prior, a second bitstream including a compressed version of the updated model parameters. The process can include outputting the first bitstream and the second bitstream for transmission to a receiver.

Instance-adaptive image and video compression using machine learning systems

A method of image compression includes receiving an image. Multiple quantized latent representations are generated to represent features of the image. Each of the quantized latent representations has a different resolution and is generated at staggered timings. Each of the later generated quantized latent representations is conditioned on each of the prior generated quantized latent representations. The multiple quantized latent representations are decoded to reconstruct the image.

Growing community of inventors

Amsterdam, Netherlands

Taco Sebastiaan Cohen