Block-wise entropy coding method in neural image compression

A system comprising:

• at least one memory configured to store computer program code;
• and
• at least one processor configured to access the computer program code and operate as instructed by the computer program code, the computer program code comprising encoding code that is configured to cause the at least one processor to:
• partition an image, from a bitstream, into a plurality of blocks;
• compress each of the plurality of blocks by a neural network-based encoder;
• obtain compressed features by obtaining a compressed feature for each block from among the plurality of blocks in the image;
• process the compressed features by an entropy encoder to generate a first compressed bitstream;
• obtain a plurality of reshaped compressed features by concatenating the compressed features such that a number of reshaped compressed features is less than a number of compressed features, wherein the concatenating the compressed features comprises concatenating the compressed feature blocks into reshaped compressed feature blocks that are larger than the compressed feature blocks, wherein a first reshaped compressed feature block has a square shape and a second reshaped feature block has a rectangular shape, wherein at least two of the reshaped compressed feature blocks are a different size from each other;
• process the plurality of reshaped compressed features by the entropy encoder to generate a second compressed bitstream;
• and
• encode the bitstream including the image based on the second compressed bitstream.

A non-transitory computer-readable medium storing computer code that is configured to, when executed by at least one processor, cause the at least one processor to implement a video encoder that:

• receives a bitstream that includes an image;
• partitions the image into a plurality of blocks;
• compresses each of the plurality of blocks by a neural network-based encoder;
• obtains compressed features by obtaining a compressed feature for each block from among the plurality of blocks in the image;
• processes the compressed features by an entropy encoder to generate a first compressed bitstream;
• obtains a plurality of reshaped compressed features by concatenating the compressed features such that a number of reshaped compressed features is less than a number of compressed features, wherein the concatenating the compressed features comprises concatenating the compressed feature blocks into reshaped compressed feature blocks that are larger than the compressed feature blocks, wherein a first reshaped compressed feature block has a square shape and a second reshaped feature block has a rectangular shape, wherein at least two of the reshaped compressed feature blocks are a different size from each other;
• processes the plurality of reshaped compressed features by the entropy encoder to generate a second compressed bitstream;
• and
• encodes the bitstream including the image based on the second compressed bitstream.