Nvidia Corporation

Partha Sriram

Ravi Prasad Bulusu

Rohit Puri

Ahmet Karakas

Zhiding Yu

Anil Ubale

Subhashree Radhakrishnan

Robert Quan

Bhagawan Reddy Gnanapa

Seunghwan Cha

In various examples, training methods as described to generate a trained neural network that is robust to various environmental features. In an embodiment, training includes modifying images of a dataset and generating boundary boxes and/or other segmentation information for the modified images which is used to train a neural network.

Automatic labeling and segmentation using machine learning models

In some embodiments, a data processing system including an operation unit including circuitry configurable to perform any selected one of a number of operations on data (e.g., audio data) and a configuration unit configured to assert configuration information to configure the operation unit to perform the selected operation. When the operation includes matrix multiplication of a data vector and a matrix whose coefficients exhibit symmetry, the configuration information preferably includes bits that determine signs of all but magnitudes of only a subset of the coefficients. When the operation includes successive addition and subtraction operations on operand pairs, the configuration information preferably includes bits that configure the operation unit to operate in an alternating addition/subtraction mode to perform successive addition and subtraction operations on each pair of data values of a sequence of data value pairs.

Configurable system for performing repetitive actions

Methods and systems for transcoding input audio data in a first encoding format to generate audio data in a second encoding format, and filterbanks for use in such systems. Some such systems include a combined synthesis and analysis filterbank (configured to generate transformed frequency-band coefficients indicative of at least one sample of the input audio data by transforming frequency-band coefficients in a manner equivalent to upsampling the frequency-band coefficients and filtering the resulting up-sampled values to generate the transformed frequency-band coefficients, where the frequency-band coefficients are partially decoded versions of input audio data that are indicative of the at least one sample) and a processing subsystem configured to generate transcoded audio data in the second encoding format in response to the transformed frequency-band coefficients. Some such methods include the steps of: generating frequency-band coefficients indicative of at least one sample of input audio data by partially decoding frequency coefficients of the input audio data; generating transformed frequency-band coefficients indicative of the at least one sample of the input audio data by transforming the frequency-band coefficients in a manner equivalent to upsampling the frequency-band coefficients to generate up-sampled values and filtering the up-sampled values; and in response to the transformed frequency-band coefficients, generating the transcoded audio data so that the transcoded audio data are indicative of each sample of the input audio data.

Method and system for efficient transcoding of audio data

A method and system for performing a 2D transform is disclosed. The 2D transform may include a row transform and/or a column transform. When performing the row or column transform, it may be determined whether each of different subsets of the data values including a partition of a row or column includes at least one zero value, whether each of different subsets of a first subset of the partition includes at least one zero value, and whether each of different subsets of at least one other subset of the partition includes at least one zero value. When performing the row or column transform, at least one transformation operation on at least one zero value may be bypassed or performed in a reduced-power manner, where such transformation operation would otherwise be performed in a manner consuming full power if the zero value were a non-zero value.

Method and system for performing two-dimensional transform on data value array with reduced power consumption

In some embodiments, a method for performing and a system configured to perform a 2D transform (for example, an inverse discrete cosine transform) on each block of a sequence of data blocks, where the 2D transform includes a row transform and a column transform. To perform the row or column transform on a row or column of data, these embodiments determine whether each of different subsets of the data values comprising a partition of the row (column) includes at least one zero value, whether each of different subsets of a first subset of the partition includes at least one zero value, and whether each of different subsets of at least one other subset of the partition includes at least one zero value. When performing the row (column) transform on each row or column that includes at least one zero value and at least one non-zero value, at least one transformation operation on at least one zero value is bypassed or performed in a reduced-power manner, where such transformation operation would otherwise be performed in a manner consuming full power if the zero value were a non-zero value. In some embodiments, the system is a pipelined video decoding system or other video processing system (or a video processing subsystem of a portable media player or other system) including a transform engine implemented in accordance with the invention. Other aspects are transform engines and transform engine circuitry for use in any embodiment of the system.

Systems and methods for optimizing system performance in variable length decoding systems are described. Embodiments are described in which decode tables are analyzed and elements of the tables sorted by probability of occurrence. Storage of elements can be determined by probability of occurrence and embodiments of the present invention can optimize system efficiency by storing most likely entries into fast-memory and least likely entries in slowest memory. In certain embodiments, a single large table is provided that cannot fit into decoder fast-memory. In some embodiments, individual elements can be optimized for storage in fast-memory by selecting more frequently occurring entries or groups of entries into decoder memory.

Optimization of decoder memory usage for VLC tables

Methods and systems that leverage the advantages of Huffman coding to increase processing efficiency of a data-stream while simultaneously minimizing storage requirements are provided. Decoding efficiency and table storage requirements can be balanced to produce systems that can be adapted for use in high-end network infrastructure applications and for low-resourced portable consumer devices. The systems and methods are operative in decoding data streams using multi-symbol codes and sign information, including AAC and MP3 data streams. A hierarchical structure of tables is described as having primary tables, secondary tables, tertiary tables and so on. Optimization balances processing requirements, table storage requirements and the described systems and methods may be implemented on a variety of processing platforms.

Growing community of inventors

Los Altos, CA, United States of America

Partha Sriram