Computationally efficient and bitrate scalable soft vector quantization
In some aspects, the techniques described herein relate to a method including: obtaining data to be compressed; determining a distance between the data to be compressed and each codeword of a plurality of codewords; selecting a predetermined number of codewords of the plurality of codewords based on the distance between the data to be compressed and each of the predetermined number of codewords; and generating compressed data, where the compressed data includes an indication of the predetermined number of codewords of the plurality of codewords.
1. A method comprising:
obtaining a stream of input vectors to be compressed;
for each of the input vectors:
determining a distance between the input vector to be compressed and each codeword of a plurality of codewords; and
selecting two or more codewords of the plurality of codewords based on the distance between the input vector to be compressed and each of the two or more codewords; and
generating compressed data that comprises an indication of the two or more codewords of the plurality of codewords.
2. The method of claim 1 , wherein generating the compressed data further comprises determining a weight associated with each of the two or more codewords.
3. The method of claim 2 , wherein generating the compressed data comprises normalizing weights associated with each of the two or more codewords to 1.
4. The method of claim 1 , further comprising providing an index for each codeword of the two or more codewords to a receive side, wherein each index indicates a codeword in a codebook at the receive side.
5. The method of claim 1 , further comprising providing an index for each codeword of the two or more codewords to a receive side as forward error correction data.
6. The method of claim 1 , wherein generating the compressed data comprises generating a weight for at least one codeword of the two or more codewords.
7. The method of claim 1 , wherein generating the compressed data further comprises determining a residual value between the compressed data and data indicated by the two or more codewords.
8. The method of claim 1 , wherein obtaining the stream of input vectors to be compressed comprises obtaining residual data from a vector quantization compression stage.
9. The method of claim 1 , wherein obtaining the stream of input vectors to be compressed comprises obtaining a subspace of a product quantization system.
10. The method of claim 1 , wherein selecting the two or more codewords of the plurality of codewords comprises selecting the two or more codewords based upon an intended bitrate for the stream of input vectors to be compressed.
11. A method comprising:
obtaining a stream of compressed data elements; and
for each of the compressed data elements:
determining a plurality of codewords from the compressed data element;
determining a weight for each of the plurality of codewords; and
determining a decompressed data element from the plurality of codewords and the weight for each of the plurality of codewords.
12. The method of claim 11 , wherein determining the plurality of codewords comprises determining a codebook index for each of the plurality of codewords from the compressed data element.
13. The method of claim 11 , wherein determining the weight for each of the plurality of codewords comprises determining weights from the compressed data element.
14. The method of claim 11 , wherein determining the weight for each of the plurality of codewords comprises determining weights for all but one of the plurality of codewords from the compressed data element and determining the weight for the one of the plurality of codewords from the weights for all but one of the plurality of codewords.
15. The method of claim 11 , wherein determining the weight for each of the plurality of codewords is based upon operation of a codec used in compression of data to produce the compressed data element.
16. One or more tangible, non-transitory computer readable mediums encoded with instructions, wherein the instructions, when executed by one or more processors, are operable to:
obtain a stream of input vectors to be compressed;
for each of the input vectors:
determine a distance between the input vector to be compressed and each codeword of a plurality of codewords; and
select two or more codewords of the plurality of codewords based on the distance between the input vector to be compressed and each of the two or more codewords; and
generate compressed data that comprises an indication of the two or more codewords of the plurality of codewords.
17. The one or more tangible, non-transitory computer readable mediums of claim 16 , wherein the instructions operable to generate the compressed data comprise instructions operable to determine a weight associated with each of the two or more codewords.
18. The one or more tangible, non-transitory computer readable mediums of claim 16 , wherein the instructions operable to generate the compressed data comprise instructions operable to determine a residual value between the compressed data and data indicated by the two or more codewords.
19. The one or more tangible, non-transitory computer readable mediums of claim 16 , wherein the instructions operable to obtain the stream of input vectors to be compressed comprise instructions operable to obtain residual data from a vector quantization compression stage.
20. The one or more tangible, non-transitory computer readable mediums of claim 16 , wherein the instructions operable to obtain the stream of input vectors to be compressed comprise instructions operable to obtain a subspace of a product quantization system.