Abstract: A device may be configured to improve object detection in reconstructed feature data according to one or more of the techniques described herein.

Abstract: A device may be configured to compress feature data according to one or more of the techniques described herein. In one example, feature data may be compressed by using residual encoding to enhance the feature data by removing redundancies. Feature data may include reshaped feature data. Enhanced feature data may be spatially down sampled and the number of channels of the enhanced feature data may be reduced by applying a 2D convolution operation. A heatmap based on the reduced enhanced feature data may be generated. The reduced enhanced feature data may be scaled using the generated heatmap. The scaled reduced enhanced feature data may be entropy encoded to generate a bitstream.

Abstract: A device may be configured to reduce distortion in compressed feature data according to one or more of the techniques described herein. In one example, a bitstream including compressed feature data may be decoded according to the video coding standard. A quantization parameter or target bit rate and a picture type may be determined for a decoded picture corresponding to a channel. A distortion reduction engine may be selected based on the quantization parameter and the picture type. The distortion reduction engine may be applied to reduce distortion.

Abstract: A method of compressing feature data includes: receiving feature data; performing spatial down sampling on the received feature data by applying a pixel unshuffle operation; and performing channel reduction on the spatially down sampled feature data by applying a non-linear two dimensional convolution with an activation.

Abstract: A device may be configured to compress feature data according to one or more of the techniques described herein. In one example, feature data may be compressed by using residual encoding to enhance the feature data by removing redundancies. Enhanced feature data may be spatially down sampled and the number of channels of the enhanced feature data may be reduced by applying a 2D convolution operation. A heatmap based on the reduced enhanced feature data may be generated. The reduced enhanced feature data may be scaled using the generated heatmap. The scaled reduced enhanced feature data may be entropy encoded to generate a bitstream.

Abstract: This disclosure discloses a method of compressing feature data corresponding to video data. The method comprising: generating feature data including a number of channels corresponding to a scale for each of N pictures included in video data, concatenating the generated feature data about the channel dimension, reducing the number of channels in the concatenated feature data to generate reduced concatenated feature data and encoding the reduced concatenated feature data into a bitstream.

Abstract: This disclosure discloses a method of reducing the impact of noise for object detection from reconstructed feature data. The method comprising: receiving reconstructed feature data including feature maps for multiple feature scales; performing a first convolution operation on the reconstructed feature data according to a defined region proposal network; performing a second convolution operation on the data resulting from the first convolution operation; further processing the data resulting from the second convolution operation according to the defined region proposal network to generate objectness logits and anchor deltas for each feature scale; and generating bounding box predictions based on the generated objectness logits and anchor deltas.

Abstract: This disclosure discloses a method of interpolating inference data corresponding to reconstructed feature data. The method comprising: receiving reconstructed feature data, wherein the reconstructed feature data corresponds to video data which has been temporally downsampled, generating bounding boxes for the reconstructed feature data, and interpolating bounding boxes for temporally downsampled portions of the video.

Abstract: A method of encoding data is disclosed. The method comprising: receiving a tensor including multiple channels of tensor values; quantizing a first group of channels of the multiple channels according to a first quantization function; quantizing a second group of channels of the multiple channels according to a second quantization function; generating a probability mass function for quantization index symbol values corresponding to the second group of channels, wherein the probability mass function is based on quantization index symbol values corresponding to the first group of channels; and entropy encoding the quantization index symbol values corresponding to the second group of channels based on the generated probability mass function.

Abstract: A method of encoding data includes: receiving an input data set having an arbitrary size about a height dimension and a width dimension; padding the input data set according to a padding function selected from a set of padding functions, such that a data set having a desired size about the height dimension and the width dimension is obtained; generating an output data set by performing a discrete convolution on the obtained data set; and generating a signal providing information corresponding to the output data set and the selected padding function.

Abstract: This disclosure relates to coding multi-dimensional data and more particularly to method for compressing feature data. The method comprising: receiving a tensor including multiple channels of tensor values; determining whether one or more channel of the multiple channels satisfies a condition; in the case where one or more of the channels do not satisfy the condition, pruning the one or more channels from the tensor; signaling data representing the tensor where the data does not include the one or more pruned channels; and signaling information indicating which of the one or more channels have been pruned from the tensor.

Abstract: A method of encoding data is disclosed. The method comprising: receiving a residual data set having a size specified by a number of channels dimension, a height dimension, and a width dimension; generating an intermediate data set corresponding to the residual data set; adding the intermediate data set to the residual data to generate a modified residual dataset; generating an output data set corresponding to the received residual data set by performing a discrete convolution on the modified residual data set, wherein performing a discrete convolution includes spatial down-sampling the modified residual data set according to a number of instances of kernels; and signaling the generated output data set in a bitstream.

Abstract: A method of encoding data includes: receiving an input data set having an arbitrary size about a height dimension and a width dimension; padding the input data set according to a padding function selected from a set of padding functions, such that a data set having a desired size about the height dimension and the width dimension is obtained; generating an output data set by performing a discrete convolution on the obtained data set; and generating a signal providing information corresponding to the output data set and the selected padding function.

Abstract: A device may be configured to compress feature data according to one or more of the techniques described herein. In one example, feature data may be compressed by using residual encoding to enhance the feature data by removing redundancies. Enhanced feature data may be spatially down sampled and the number of channels of the enhanced feature data may be reduced by applying a 2D convolution operation. A heatmap based on the reduced enhanced feature data may be generated. The reduced enhanced feature data may be scaled using the generated heatmap. The scaled reduced enhanced feature data may be entropy encoded to generate a bitstream.

Abstract: Methods of encoding and decoding for video data are described for encoding or decoding multi-level significance maps. Distinct context sets may be used for encoding the significant-coefficient flags in different regions of the transform unit. In a fixed case, the regions are defined by coefficient group borders. In one example, the upper-left coefficient group is a first region and the other coefficient groups are a second region. In a dynamic case, the regions are defined by coefficient group borders, but the encoder and decoder dynamically determine in which region each coefficient group belongs. Coefficient groups may be assigned to one region or another based on, for example, whether their respective significant-coefficient-group flags were inferred or not.

Abstract: Methods of encoding and decoding for video data are described for encoding or decoding multi-level significance maps. Distinct context sets may be used for encoding the significant-coefficient flags in different regions of the transform unit. In a fixed case, the regions are defined by coefficient group borders. In one example, the upper-left coefficient group is a first region and the other coefficient groups are a second region. In a dynamic case, the regions are defined by coefficient group borders, but the encoder and decoder dynamically determine in which region each coefficient group belongs. Coefficient groups may be assigned to one region or another based on, for example, whether their respective significant-coefficient-group flags were inferred or not.

Abstract: A method of compressing feature data includes: receiving feature data; performing spatial down sampling on the received feature data by applying a pixel unshuffle operation; and performing channel reduction on the spatially down sampled feature data by applying a non-linear two dimensional convolution with an activation.

Abstract: Methods of encoding and decoding for video data are described in which multi-level significance maps are used in the encoding and decoding processes. The significant-coefficient flags that form the significance map are grouped into contiguous groups, and a significant-coefficient-group flag signifies for each group whether that group contains no non-zero significant-coefficient flags. If there are no non-zero significant-coefficient flags in the group, then the significant-coefficient-group flag is set to zero. The set of significant-coefficient-group flags is encoded in the bitstream. Any significant-coefficient flags that fall within a group that has a significant-coefficient-group flag that is non-zero are encoded in the bitstream, whereas significant-coefficient flags that fall within a group that has a significant-coefficient-group flag that is zero are not encoded in the bitstream.

Abstract: A method of encoding data includes: receiving an input data set having an arbitrary size about a height dimension and a width dimension; padding the input data set according to a padding function selected from a set of padding functions, such that a data set having a desired size about the height dimension and the width dimension is obtained; generating an output data set by performing a discrete convolution on the obtained data set; and generating a signal providing information corresponding to the output data set and the selected padding function.

Abstract: Methods of encoding and decoding for video data are described in which multi-level significance maps are used in the encoding and decoding processes. The significant-coefficient flags that form the significance map are grouped into contiguous groups, and a significant-coefficient-group flag signifies for each group whether that group contains no non-zero significant-coefficient flags. If there are no non-zero significant-coefficient flags in the group, then the significant-coefficient-group flag is set to zero. The set of significant-coefficient-group flags is encoded in the bitstream. Any significant-coefficient flags that fall within a group that has a significant-coefficient-group flag that is non-zero are encoded in the bitstream, whereas significant-coefficient flags that fall within a group that has a significant-coefficient-group flag that is zero are not encoded in the bitstream.