Abstract: This disclosure provides methods, devices, and systems for video coding. The present implementations more specifically relate to autoencoders that support infer-frame coding in the latent domain. A video encoder may convert a frame of video from the pixel domain to the latent domain based on a machine learning model. For example, the machine learning model may be trained to transform the video frame into a tensor of latent attributes. In some aspects, the video encoder may combine the resulting tensor with a tensor of latent attributes associated with a previously-encoded video frame and transform the resulting tensor into a vector that includes latent attributes from both the current video frame and the previous video frame based on an inter-coding transform. More specifically, the inter-coding transform may reduce a dimensionality of the combined tensor so that the resulting vector is smaller or more compressible than the original tensor of latent attributes.

Abstract: Systems and methods for controlling compression of image data by a quality controller include obtaining a desired target number of bits to be generated from compression of a current image area using a predetermined compression protocol, determining a calculated quantisation level based on the desired number of bits using a predetermined relationship between the number of bits and quantisation level, selecting a discrete quantisation level from a plurality of predetermined discrete quantisation levels based on the calculated quantisation level, determining a predicted number of bits that would result from compression of the current image area at the selected discrete quantisation level using the predetermined relationship, determining whether the predicted number of bits exceeds the desired number of bits and, if not, providing to an encoder information to enable the encoder to determine a set of compression parameters associated with the selected discrete quantisation level.

Abstract: This disclosure provides methods, devices, and systems for data compression. The present implementations more specifically relate to encoding techniques for compressing probability tables used for entropy coding. In some aspects, an entropy encoder may encode a probability table so that one or more contexts are represented by fewer bits than would otherwise be needed to represent the frequency of each symbol as a proportion of the total frequency of all symbols associated with such contexts. For example, if a given row of the probability table (prior to encoding) includes a number (M) of entries each having a binary value represented by a number (K) of bits, the same row of entries may be represented by fewer than M*K bits in the encoded probability table.

Abstract: This disclosure provides methods, devices, and systems for image scaling. The present implementations more specifically relate to downscaling techniques that preserve a subset of pixel values in a digital image so that all reconstructed pixel values in the upscaled image can be interpolated based on two or more preserved pixel values. In some aspects, an image downscaler may partition a digital image into a number of image tiles based on a base tile size and one or more scaling factors associated with a downscaling operation used to downscale each image tile. The base tile size indicates the dimensions of the downscaled tiles. The one or more scaling factors indicate distances between the pixel values preserved as a result of one or more iterations, respectively, of the downscaling operation, where each iteration preserves a smaller subset of the pixel values from the original image (as a lower resolution tile).

Abstract: This disclosure provides methods, devices, and systems for data compression. The present implementations more specifically relate to encoding techniques for compressing probability tables used for entropy coding. In some aspects, an entropy encoder may encode a probability table so that one or more contexts are represented by fewer bits than would otherwise be needed to represent the frequency of each symbol as a proportion of the total frequency of all symbols associated with such contexts. For example, if a given row of the probability table (prior to encoding) includes a number (M) of entries each having a binary value represented by a number (K) of bits, the same row of entries may be represented by fewer than M*K bits in the encoded probability table.

Abstract: This disclosure provides methods, devices, and systems for data compression. The present implementations more specifically relate to lossless data compression techniques that combine entropy coding with run-length encoding of zeroes. In some aspects, an encoder may obtain a sequence of integer values representing a frame of data and may iteratively scan the sequence for runs of zeros. During each iteration of the scan, the encoder may increment a count value (N) if the current integer value is equal to zero or, if the current integer value is not equal to zero, output a pattern of bits representing the current count value or the current integer value. In some implementations, the pattern of bits may include a codeword associated with an entropy encoding scheme. More specifically, the encoder may encode the current integer value as the codeword if N>0 and may encode the current count value as the codeword if N=0.

Abstract: Systems and methods for controlling compression of image data by a quality controller include obtaining a desired target number of bits to be generated from compression of a current image area using a predetermined compression protocol, determining a calculated quantisation level based on the desired number of bits using a predetermined relationship between the number of bits and quantisation level, selecting a discrete quantisation level from a plurality of predetermined discrete quantisation levels based on the calculated quantisation level, determining a predicted number of bits that would result from compression of the current image area at the selected discrete quantisation level using the predetermined relationship, determining whether the predicted number of bits exceeds the desired number of bits and, if not, providing to an encoder information to enable the encoder to determine a set of compression parameters associated with the selected discrete quantisation level.