Abstract: A method includes accessing a plurality of weight matrices of a machine learning model. The method also includes, for each weight matrix, decomposing the weight matrix into a U matrix, an S matrix, and a V matrix using singular value decomposition. The S matrix is a diagonal matrix, and a singular group corresponds to each element in the S matrix. The method further includes, for each weight matrix, determining an importance score of each singular group. The importance score of the singular group represents a change in loss if the singular group is removed from the machine learning model. The method also includes, for each weight matrix, ranking the singular groups across the plurality of weight matrices based on the importance scores. In addition, the method includes, for each weight matrix, identifying one or more of the singular groups to prune based on the ranking of the singular groups.

Abstract: In one embodiment, a method includes accessing at least a portion of a training dataset for a trained neural network that includes multiple layers, where each layer includes a number of parameters, and where the training dataset includes multiple training samples that each include an input and a ground-truth output used to train the trained neural network. The method further includes training a hypernetwork to generate a layer-specific compression mask for each of one or more of the multiple layers of the trained neural network. The method further includes generating, by the trained hypernetwork, a final layer-specific compression mask for the trained neural network and compressing the trained neural network by reducing, for each of the one or more layers of the neural network, the number of parameters of that layer according to the final layer-specific compression mask.

Abstract: A method and an apparatus for computing a parallax are provided. The method includes: obtaining a plurality of views of a pixel point; computing gradients in a plurality of directions of the pixel point in the plurality of view; computing a matching error corresponding to a parallax of the pixel point according to the gradients; and computing the parallax of the current pixel point according to the matching error. Through the method and the apparatus for computing a parallax, the accuracy of parallax estimation is increased.

Abstract: A method for measuring multimedia communication quality is disclosed. The multimedia video communication quality may be objectively reflected through the embedment and extraction of digital watermark under a precondition that the quality of the multimedia video data is not obviously affected. In the invention, each frame of the multimedia video data is uniformly divided into blocks of equal size and watermark data is embedded in each of the blocks, so that the watermark may be uniformly distributed. The multimedia video data are divided into groups, and the watermark is embedded in a part of the frames with equal interval between the frames in each group to reduce the effect of the watermark on the data. The watermark information is directly embedded in the spatial domain of the original video data.

Abstract: A method and an apparatus for computing a parallax are provided. The method includes: obtaining a plurality of views of a pixel point; computing gradients in a plurality of directions of the pixel point in the plurality of view; computing a matching error corresponding to a parallax of the pixel point according to the gradients; and computing the parallax of the current pixel point according to the matching error. Through the method and the apparatus for computing a parallax, the accuracy of parallax estimation is increased.

Abstract: The present invention discloses a method for recovering a lost data unit, the method including: partitioning data to be transmitted into one or more data units, sorting the data units according to importance levels of the data units, and determining a check rule for the one or more data units, performing, by a transmitting end, a calculation on the sorted data units according to a predetermined algorithm depending on the check rule, to generate one or more corresponding check units; transmitting, by the transmitting end, the one or more data units to a receiving end in a sorted order, and transmitting the corresponding check units to the receiving end; recovering, by the receiving end, a lost data unit according to the receiving data units and check units as well as the check rule. With the method, the loss-preventative capability of the data units may be improved.

Abstract: A video coding method based on a motion skip mode (MSM) is provided. The method includes the following steps. A corresponding reference block of a current macro block to be encoded in a view-point reference image is determined, according to a direction of a disparity vector from a current image relative to the view-point reference image deduced by using a block smaller than 16×16 pixels as a base unit. The current macro block to be encoded is then encoded according to motion information of a macro block that the determined corresponding reference block belongs to. Other related video coding methods and corresponding codecs based on the MSM are also provided. Therefore, macro block motion information (MMI) of the currently encoded macro block at a corresponding position in the view-point reference image can be more accurately obtained, thereby improving a coding efficiency of the MSM.

Abstract: The present invention discloses a method for acquiring automatic exposure control parameters and a method for controlling automatic exposure control parameters and an imaging device. The core idea is to take the brightness weighted average value obtained according to brightness distribution weighted statistics of the image as the exposure control parameter. Then through adjusting the weighted coefficient of pixels in different brightness regions, control the influence of the pixel concentrated brightness regions on the brightness average value. Because the brightness distribution is the basis for determining weighted statistics, the control method of the present invention is not influenced by the scene assumption. Rather it can be universally adapted to a variety of different scenes, and the exposure control effect is not influenced by the position of scenes in the actual bright region/dark region and block distribution.

Abstract: A method for concealing errors includes that the transmitting end splits the received compressed video data into slice structures, allocates the adjacent slice structures to non-adjacent OFDM sub-channels or sub-channel groups, and sends the slice structures to the receiving end; the receiving end conceals the errors on a slice structure according to the slice structure which is chronologically or spatially related to the erroneous slice structure if detecting that any error occurs on the slice structure. The transmitting end includes a slice splitting module and a frequency domain interleaving module; the receiving end includes a domain de-interleaving module, a decompression and error detecting module, and an error concealing module.

Abstract: A method for measuring multimedia communication quality is disclosed. The multimedia video communication quality may be objectively reflected through the embedment and extraction of digital watermark under a precondition that the quality of the multimedia video data is not obviously affected. In the invention, each frame of the multimedia video data is uniformly divided into blocks of equal size and watermark data is embedded in each of the blocks, so that the watermark may be uniformly distributed. The multimedia video data are divided into groups, and the watermark is embedded in a part of the frames with equal interval between the frames in each group to reduce the effect of the watermark on the data. The watermark information is directly embedded in the spatial domain of the original video data.

Abstract: A method for protecting video transmission based on H.264 is disclosed, In the invention, a receiving end performs statistical analysis on the communication quality and the result is fed back to a transmitting end, the transmitting end adjusts the protection strategy self adaptively; the receiving end performs statistical analysis on the accumulated lost slice rate and position information about lost data based on the discontinuous situation of NALU sequence numbers; the receiving end feeds the statistical information back to the transmitting end with extended SEI message; the transmitting end configures two sets of protection strategy systems to separately protect critical data and non critical data, and selects the protection strategy of appropriate level according to the accumulated lost slice rate that is fed back; the receiving end transmits position information about lost data back to the transmitting end; the transmitting end resends the lost data.

Abstract: The invention relates to a method for eliminating errors in compressed video transmission. A method for eliminating errors in H.264 compressed video transmission is disclosed, to eliminating errors in a simple and efficient manner. In the invention, an error information feedback mechanism is used in combination with error concealment and error propagation suppression to eliminate errors. An error is detected and statistics of error information, such as position of lost data, is obtained by sequence numbers of NALUs and information for carrying slices. An error information feedback channel is established in H.264 architecture by defining an extended SEI message. The error propagation is suppressed by segment-wise intra-coding in batches.

Abstract: The present invention discloses a method for recovering a lost data unit, the method including: partitioning data to be transmitted into one or more data units, sorting the data units according to importance levels of the data units, and determining a check rule for the one or more data units; performing, by a transmitting end, a calculation on the sorted data units according to a predetermined algorithm depending on the check rule, to generate one or more corresponding check units; transmitting, by the transmitting end, the one or more data units to a receiving end in a sorted order, and transmitting the corresponding check units to the receiving end; recovering, by the receiving end, a lost data unit according to the received data units and check units as well as the check rule. With the method, the loss-preventive capability of the data units may be improved.

Abstract: The present invention discloses a method for acquiring automatic exposure control parameters and a method for controlling automatic exposure control parameters and an imaging device. The core idea is to take the brightness weighted average value obtained according to brightness distribution weighted statistics of the image as the exposure control parameter. Then through adjusting the weighted coefficient of pixels in different brightness regions, control the influence of the pixel concentrated brightness regions on the brightness average value. Because the brightness distribution is the basis for determining weighted statistics, the control method of the present invention is not influenced by the scene assumption. Rather it can be universally adapted to a variety of different scenes, and the exposure control effect is not influenced by the position of scenes in the actual bright region/dark region and block distribution.