Abstract: Techniques for enhancing machine learning (ML) model execution. The technique includes determining an amount of memory used to process layers of a machine learning network having multiple layers, smoothing the amount of memory used to process the layers of the machine learning network based on a number of layers, identifying change layers where the smoothed amount of memory used changes more than a memory change threshold amount, grouping the layers of the machine learning network into a first layer grouping based on the identified change layers, and outputting the first layer grouping.

Abstract: An apparatus includes first CPU and second CPU cores, a L1 cache subsystem coupled to the first CPU core and comprising a L1 controller, and a L2 cache subsystem coupled to the L1 cache subsystem and to the second CPU core. The L2 cache subsystem includes a L2 memory and a L2 controller configured to operate in an aliased mode in response to a value in a memory map control register being asserted. In the aliased mode, the L2 controller receives a first request from the first CPU core directed to a virtual address in the L2 memory, receives a second request from the second CPU core directed to the virtual address in the L2 memory, directs the first request to a physical address A in the L2 memory, and directs the second request to a physical address B in the L2 memory.

Abstract: A method for convolution in a convolutional neural network (CNN) is provided that includes accessing a coefficient value of a filter corresponding to an input feature map of a convolution layer of the CNN, and performing a block multiply accumulation operation on a block of data elements of the input feature map, the block of data elements corresponding to the coefficient value, wherein, for each data element of the block of data elements, a value of the data element is multiplied by the coefficient value and a result of the multiply is added to a corresponding data element in a corresponding output block of data elements comprised in an output feature map.

Abstract: A matching accelerator in the form of a hardware accelerator configured to perform matrix multiplication and/or additional operations is used to optimize keypoint matching. An SSE calculation may be determined by utilizing the matching 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.

Abstract: Several methods and systems for facilitating multimedia data encoding are disclosed. In an embodiment, a plurality of picture buffers associated with multimedia data are received in an order of capture associated with the plurality of picture buffers. Buffer information is configured for each picture buffer from among the plurality of picture buffers comprising at least one of a metadata associated with the corresponding picture buffer and one or more encoding parameters for the corresponding picture buffer. A provision of picture buffers in an order of encoding is facilitated based on the configured buffer information.

Abstract: 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.

Abstract: A method for convolution in a convolutional neural network (CNN) is provided that includes accessing a coefficient value of a filter corresponding to an input feature map of a convolution layer of the CNN, and performing a block multiply accumulation operation on a block of data elements of the input feature map, the block of data elements corresponding to the coefficient value, wherein, for each data element of the block of data elements, a value of the data element is multiplied by the coefficient value and a result of the multiply is added to a corresponding data element in a corresponding output block of data elements comprised in an output feature map.

Abstract: Several methods and systems for facilitating multimedia data encoding are disclosed. In an embodiment, a plurality of picture buffers associated with multimedia data are received in an order of capture associated with the plurality of picture buffers. Buffer information is configured for each picture buffer from among the plurality of picture buffers comprising at least one of a metadata associated with the corresponding picture buffer and one or more encoding parameters for the corresponding picture buffer. A provision of picture buffers in an order of encoding is facilitated based on the configured buffer information.

Abstract: A method for dynamically quantizing feature maps of a received image. The method includes convolving an image based on a predicted maximum value, a predicted minimum value, trained kernel weights and the image data. The input data is quantized based on the predicted minimum value and predicted maximum value. The output of the convolution is computed into an accumulator and re-quantized. The re-quantized value is output to an external memory. The predicted min value and the predicted max value are computed based on the previous max values and min values with a weighted average or a pre-determined formula. Initial min value and max value are computed based on known quantization methods and utilized for initializing the predicted min value and predicted max value in the quantization process.

Abstract: In accordance with disclosed embodiments, an image processing method includes performing a first scan in a first direction on a first list of pixels in which, for each pixel in the first list, a feature point property is compared with a corresponding feature point property of each of a first set of neighboring pixels, performing a second scan in a second direction on the first list of pixels in which, for each pixel in the first list, a feature point property is compared with a corresponding feature point property of each of a second set of neighboring pixels, using the results of the first and second scans to identify pixels from the first list to be suppressed, and forming a second list of pixels that includes pixels from the first list that are not identified as pixels to be suppressed. The second list represents a non-maxima suppressed list.

Abstract: A method for object classification in a decision tree based adaptive boosting (AdaBoost) classifier implemented on a single-instruction multiple-data (SIMD) processor is provided that includes receiving feature vectors extracted from N consecutive window positions in an image in a memory coupled to the SIMD processor and evaluating the N consecutive window positions concurrently by the AdaBoost classifier using the feature vectors and vector instructions of the SIMD processor, in which the AdaBoost classifier concurrently traverses decision trees for the N consecutive window positions until classification is complete for the N consecutive window positions.

Abstract: A method of determining a summation of pixel characteristics for a rectangular region of a digital image includes determining if a base address for a data element in an integral image buffer is aligned for an SIMD operation by a processor embedded in an electronic assembly configured to perform Haar-like feature calculations. The data element represents a corner of the rectangular region of an integral image. The integral image is a representation of the digital image. The integral image is formed by data elements stored in the integral image buffer. The data element is loaded from the integral image buffer to the processor when the base address is aligned for the SIMD operation. An offset data element of an offset integral image is loaded from an offset integral buffer when the base address is non-aligned for the SIMD operation. The offset data element represents the corner of the rectangular region.

Abstract: A method for video object detection includes detecting an object in a first video frame, and selecting a first interest point and a second interest point of the object. The first interest point is in a first region of interest located at a first corner of a box surrounding the object. The second interest point is in a second region of interest located at a second corner of the box. The second corner is diagonally opposite the first corner. A first optical flow of the first interest point and a second optical flow of the second interest point are determined. A location of the object in a second video frame is estimated by determining, in the second video frame, a location of the first interest point based on the first optical flow and a location of the second interest point based on the second optical flow.

Abstract: A method for dynamically quantizing feature maps of a received image. The method includes convolving an image based on a predicted maximum value, a predicted minimum value, trained kernel weights and the image data. The input data is quantized based on the predicted minimum value and predicted maximum value. The output of the convolution is computed into an accumulator and re-quantized. The re-quantized value is output to an external memory. The predicted min value and the predicted max value are computed based on the previous max values and min values with a weighted average or a pre-determined formula. Initial min value and max value are computed based on known quantization methods and utilized for initializing the predicted min value and predicted max value in the quantization process.

Abstract: 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.

Abstract: A method for video object detection includes detecting an object in a first video frame, and selecting a first interest point and a second interest point of the object. The first interest point is in a first region of interest located at a first corner of a box surrounding the object. The second interest point is in a second region of interest located at a second corner of the box. The second corner is diagonally opposite the first corner. A first optical flow of the first interest point and a second optical flow of the second interest point are determined. A location of the object in a second video frame is estimated by determining, in the second video frame, a location of the first interest point based on the first optical flow and a location of the second interest point based on the second optical flow.

Abstract: 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.

Abstract: Techniques including receiving a first set of values for processing by a machine learning (ML) network, storing a first portion of the first set of values in an on-chip memory, processing the first portion of the first set of values in a first layer of the ML network to generate a second portion of a second set of values, overwriting the stored first portion with the generated second portion, processing the second portion in a second layer of the ML network to generate a third portion of a third set of values, storing the third portion, repeating the steps of storing the first portion, processing the first portion, overwriting the stored first portion, processing the second portion, and storing the third portion for a fourth portion of the first set of values until all portions of the first set of values are processed to generate the third set of values.

Abstract: Several techniques aimed at reducing computational complexity when encoding uses bi-predictively encoded frames (B-frames) are implemented in a video encoder. In an embodiment, B-frames are not used as reference frames for encoding P-frames and other B-frames. Non-use of B-frames allows a de-blocking filter used in the video encoder to be switched off when reconstructing encoded B-frames, and use of a lower complexity filter for fractional-resolution motion search for B-frames. In another embodiment, cost functions used in motion estimation for B-frames are simplified to reduce computational complexity. In one more embodiment, fractional pixel refinement in motion search for B-frames is simplified. In yet another embodiment, predictors used in motion estimation for a macro-block in a P-frame are selected from a B-frame that uses a same reference frame as the P-frame.

Abstract: Several techniques aimed at reducing computational complexity when encoding uses bi-predictively encoded frames (B-frames) are implemented in a video encoder. In an embodiment, B-frames are not used as reference frames for encoding P-frames and other B-frames. Non-use of B-frames allows a de-blocking filter used in the video encoder to be switched off when reconstructing encoded B-frames, and use of a lower complexity filter for fractional-resolution motion search for B-frames. In another embodiment, cost functions used in motion estimation for B-frames are simplified to reduce computational complexity. In one more embodiment, fractional pixel refinement in motion search for B-frames is simplified. In yet another embodiment, predictors used in motion estimation for a macro-block in a P-frame are selected from a B-frame that uses a same reference frame as the P-frame.