Texas Instruments Corporation

Soyeb Noormohammed Nagori

Manu Mathew

Arun Shankar Kudana

Naveen Srinivasamurthy

Hrushikesh Tukaram Garud

Deepak Kumar Poddar

Pramod Kumar Swami

Manoj Koul

Mihir Narendra Mody

Prashanth Ramanathpur Viswanath

Mahant Siddaramanna

Anurag Mithalal Jain

Shyam Jagannathan

Kedar Satish Chitnis

Peter Labaziewicz

Debapriya Maji

Kumar Arrakutti Desappan

Uday Pudipeddi Kiran

Poorna Kumar

Ankit Ajmani

Ajit Venkat Rao

Kumar Desappan

Do-Kyoung Kwon

Akira Osamoto

Dipan Kumar Mandal

Satish Kumar

Zhan Ma

Girish Srinivasa Murthy

Mudit Mehrotra

Prashanth Ramanathpu Viswanath

A hardware accelerator is configured to perform matrix multiplication and/or additional operations to optimize keypoint matching. A sum of squared error (SSE) calculation may be determined by utilizing the hardware accelerator to perform matrix multiplication to obtain a cost matrix for two sets of keypoint descriptors from two images. The hardware accelerator may determine a best cost calculation for each keypoint in each direction, which is utilized to perform keypoint matching.

Two-way descriptor matching on deep learning accelerator

Several methods and systems for estimating motion in a plurality of multimedia pictures are disclosed. In an embodiment, at least one temporal distance between a multimedia picture and one or more reference pictures corresponding to the multimedia picture in a capture order associated with the plurality of multimedia pictures is computed. The at least one temporal distance is computed subsequent to an encoding of the multimedia picture. At least one motion estimation parameter is determined based on the at least one temporal distance. Motion associated with a subsequent multimedia picture to be encoded is estimated based on the at least one motion estimation parameter.

Methods and systems for temporal distance based estimating of motion in multimedia picture encoding based on previously encoded predictive picture

A method for multi-label image classification in a convolutional neural network (CNN) is provided that includes forming a composite image from a plurality of clipped images, and processing the composite image by the CNN to generate a probability vector for each clipped image of the plurality of clipped images, wherein a length of a probability vector is equal to a number of classes the CNN is designed to classify.

Multi-label image classification in a deep learning network

A method of rate control in coding of a video sequence to generate a compressed bit stream is provided that includes computing a sequence base quantization step size for a sequence of pictures in the video sequence, computing a picture base quantization step size for a picture in the sequence of pictures based on the sequence base quantization step size, a type of the picture, and a level of the picture in a rate control hierarchy, and coding the picture using the picture base quantization step size to generate a portion of the compressed bit stream.

Rate control in video coding

Various embodiments of the present technology relate to using neural networks to detect objects in images. More specifically, some embodiments relate to the reduction of computational analysis regarding object detection via neural networks. In an embodiment, a method of performing object detection is provided. The method comprises determining, via a convolution neural network, at least a classification of an image, wherein the classification corresponds to an object in the image and comprises location vectors corresponding to pixels of the image. The method also comprises, for at least a location vector of the location vectors, obtaining a confidence level, wherein the confidence level represents a probability of the object being present at the location vector, and calculating an upper-bound score based at least on the confidence level. The method further comprises, for at least an upper-bound score based at least on the confidence level, performing an activation function on the upper-bound score, and classifying, via a detection layer, the object in the image.

Efficient object detection using deep learning techniques

This invention predicts that intra mode prediction is more effective for the macroblocks where motion estimation in inter mode prediction fails. This failure is indicated by a large value of the inter mode SAD. This invention performs intra mode prediction for only macro blocks have larger inter mode SADs. The definition of a large inter mode SAD differs for different content. This invention compares the inter mode SAD of a current macroblock with an adaptive threshold. This adaptive threshold depends on the average and variance of the SADs of the previous predicted frame. An adaptive threshold is calculated for each new predictive frame.

Intra/inter mode decision for predictive frame encoding

The disclosure provides a sample adaptive offset (SAO) encoder. The SAO encoder includes a statistics collection (SC) block and a rate distortion optimization (RDO) block coupled to the SC block. The SC block receives a set of deblocked pixels and a set of original pixels. The SC block categorizes each deblocked pixel of the set of deblocked pixels in at least one of a plurality of band and edge categories. The SC block estimates an error in each category as difference between a deblocked pixel of the set of deblocked pixels and corresponding original pixel of the set of original pixels. The RDO block determines a set of candidate offsets associated with each category and selects a candidate offset with a minimum RD cost. The minimum RD cost is used by a SAO type block and a decision block to generate final offsets for the SAO encoder.

Method and apparatus for real-time SAO parameter estimation

A computer vision system is provided that includes a camera capture component configured to capture an image from a camera, a memory, and an image compression decompression engine (ICDE) coupled to the memory and configured to receive each line of the image, and compress each line to generate a compressed bit stream. To compress a line, the ICDE is configured to divide the line into compression units, and compress each compression unit, wherein to compress a compression unit, the ICDE is configured to perform delta prediction on the compression unit to generate a delta predicted compression unit, compress the delta predicted compression unit using exponential Golomb coding to generate a compressed delta predicted compression unit, and add the compressed delta predicted compression unit to the compressed bit stream.

Image compression/decompression in a computer vision system

In described examples of a method for quantizing data for a convolutional neural network (CNN) is provided. A set of data is received and quantized the using a power-of-2 parametric activation (PACT2) function. The PACT2 function arranges the set of data as a histogram and discards a portion of the data corresponding to a tail of the histogram to form a remaining set of data. A clipping value is determined by expanding the remaining set of data to a nearest power of two value. The set of data is then quantized using the clipping value. With PACT2, a model can be quantized either using post training quantization or using quantization aware training. PACT2 helps a quantized model to achieve close accuracy compared to the corresponding floating-point model.

Parametric power-of-2 clipping activations for quantization for convolutional neural networks

Several methods and systems for encoding pictures associated with video data are disclosed. In an embodiment, a method includes determining by a processing module, whether a picture is to be encoded based on at least one of a skip assessment associated with the picture and an encoding status of a pre-selected number of pictures preceding the picture in an encoding sequence. The method further includes encoding by the processing module, a plurality of rows of video data associated with the picture upon determining that the picture is to be encoded, wherein the plurality of rows are encoded based on a pre-selected maximum encoded picture size.

Methods and systems for encoding pictures associated with video data

Techniques for localizing a vehicle include obtaining an image from a camera, identifying a set of image feature points in the image, obtaining an approximate location of the vehicle, determining a set of sub-volumes (SVs) of a map to access based on the approximate location, obtaining map feature points and associated map feature descriptors associated with the set of SVs, determining a set of candidate matches between the set of image feature points and the obtained map feature points, determining a set of potential poses of the camera from candidate matches from the set of candidate matches and an associated reprojection error estimated for remaining points to select a first pose of the set of potential poses having a lowest associated reprojection error, determining the first pose is within a threshold value of an expected vehicle location, and outputting a vehicle location based on the first pose.

Camera-only-localization in sparse 3D mapped environments

A method and system for bit rate control during encoding of multimedia data are disclosed. A change in complexity of a multimedia picture relative to complexity associated with one or more multimedia pictures in a multimedia sequence is determined. A complexity associated with a multimedia picture is determined based on number of bits and an average quantization associated with the multimedia picture. A bit rate is adjusted for encoding the multimedia picture based on the change in complexity of the multimedia picture. The bit rate is increased on determining an increase in complexity of the multimedia picture and is decreased on determining a decrease in complexity of the multimedia picture. Utilization of additional bits during the increase in the bit rate and saving of bits during the decrease in the bit rate are compensated during adjusting of bit rates for encoding subsequent multimedia pictures in the multimedia sequence.

Method and system of bit rate control

A video encoder receives a minimum number of bits (MIN) and a maximum number of bits (MAX) to be used to encode a segment of a sequence of image frames, the segment including a set of pictures contained in the sequence of image frames. The video encoder encodes the set of pictures using a total number of bits greater than the minimum number of bits (MIN), and not exceeding the maximum number of bits (MAX). Thus, the transmission bit-rate of the video encoder can be constrained to lie within a maximum and minimum rate. In an embodiment, the constraints are enforced over relatively short time intervals.

Transmission bit-rate control in a video encoder

Several methods and systems for encoding of multimedia pictures are disclosed. In an embodiment, an occupancy level of a coded picture buffer (CPB) associated with a hypothetical reference decoder (HRD) is estimated at an instant of removal of an access unit corresponding to a multimedia picture from the CPB for decoding the access unit. A number of bits for encoding the multimedia picture is allocated based on the estimated occupancy level of the CPB. The multimedia picture is encoded based on the allocated number of bits.

Methods and systems for encoding of multimedia pictures

According to an aspect, a video encoder selects a block of intermediate size from a set of block sizes for intra-prediction estimation for encoding a video signal. A set of neighboring blocks with the intermediate size are tested for combining. If the set of neighboring blocks are determined to be combinable, the video encoder selects a larger block size formed by the tested neighboring blocks for encoding. On the other hand, if the set of neighboring blocks are determined to be not combinable, the video encoder selects a smaller block size from the set of tested neighboring blocks for prediction. According to another aspect of the present disclosure, the best mode for intra-prediction is determined by first intra-predicting a block with intermediate modes in a set of modes. Then the intra-predictions are performed for the neighboring modes of at least one intermediate mode.

Growing community of inventors

Bengaluru, India

Soyeb Noormohammed Nagori