Techniques for parameter set and header design for compressed neural network representation
Systems and methods for encoding and decoding neural network data is provided. A method includes: obtaining an independent neural network with a topology; encoding the independent neural network with the topology such as to obtain a neural network representation (NNR) bitstream; and sending the NNR bitstream to a decoder, wherein the NNR bitstream includes a group of NNR units (GON) that represents the independent neural network with the topology, and the GON includes an NNR model parameter set unit, an NNR layer parameter set unit, an NNR topology unit, an NNR quantization unit, and an NNR compressed data unit.
1. A method performed by at least one processor, the method comprising:
obtaining an independent neural network with a topology;
encoding the independent neural network with the topology such as to obtain a neural network representation (NNR) bitstream; and
sending the NNR bitstream to a decoder,
wherein the NNR bitstream includes a group of NNR units (GON) that represents the independent neural network with the topology, and the GON includes an NNR model parameter set unit, an NNR layer parameter set unit, an NNR topology unit, an NNR quantization unit, and an NNR compressed data unit, and
each of the NNR model parameter set unit, the NNR layer parameter set unit, the NNR topology unit, the NNR quantization unit, and the NNR compressed data unit is a respective at least one NNR unit, of the GON, that each includes a header and a payload.
2. The method of claim 1 , wherein the GON is included in one or more aggregate NNR units of the NNR bitstream, and the one or more aggregate NNR units each include an aggregate NNR unit header and an aggregate NNR unit payload, the aggregate NNR unit payload including at least a portion of the NNR units of the GON.
3. The method of claim 2 , wherein the GON is included in a single aggregate NNR unit.
4. The method of claim 3 , wherein the single aggregate NNR unit includes a syntax element that indicates a type of the single aggregate NNR unit as a self-contained NNR aggregate unit.
5. The method of claim 4 , wherein the syntax element is included in an aggregate NNR unit header of the single aggregate NNR unit.
6. The method of claim 1 , wherein the NNR model parameter set unit includes a syntax element that indicates that NNR units, that refer to the NNR model parameter set unit, are independently decodable.
7. The method of claim 6 , wherein the syntax element is included in a header of the NNR model parameter set unit.
8. The method of claim 1 , wherein the NNR layer parameter set unit includes a syntax element that indicates that NNR units, that refer to the NNR layer parameter set unit, are independently decodable.
9. The method of claim 8 , wherein the syntax element is included in a header of the NNR layer parameter set unit.
10. The method of claim 1 , wherein, in the GON, the NNR model parameter set unit is followed by the NNR layer parameter set unit.
11. A system comprising:
at least one processor; and
memory storing computer code, the computer code comprising:
encoding code configured to cause the at least one processor to encode an independent neural network with a topology such as to obtain a neural network representation (NNR) bitstream, and
sending code configured to cause the at least one processor to send the NNR bitstream to a decoder;
wherein the NNR bitstream includes a group of NNR units (GON) that represents the independent neural network with the topology, and the GON includes an NNR model parameter set unit, an NNR layer parameter set unit, an NNR topology unit, an NNR quantization unit, and an NNR compressed data unit, and
each of the NNR model parameter set unit, the NNR layer parameter set unit, the NNR topology unit, the NNR quantization unit, and the NNR compressed data unit is a respective at least one NNR unit, of the GON, that each includes a header and a payload.
12. The system of claim 11 , wherein the GON is included in one or more aggregate NNR units of the NNR bitstream, and the one or more aggregate NNR units each include an aggregate NNR unit header and an aggregate NNR unit payload, the aggregate NNR unit payload including at least a portion of the NNR units of the GON.
13. The system of claim 12 , wherein the GON is included in a single aggregate NNR unit.
14. The system of claim 13 , wherein the single aggregate NNR unit includes a syntax element that indicates a type of the single aggregate NNR unit as a self-contained NNR aggregate unit, and the computer code comprises determining code that is configured to determine that the single aggregate NNR unit is self-contained based on the syntax element.
15. The system of claim 14 , wherein the syntax element is included in an aggregate NNR unit header of the single aggregate NNR unit.
16. The system of claim 11 , wherein the NNR model parameter set unit includes a syntax element that indicates that NNR units, that refer to the NNR model parameter set unit, are independently decodable, and the computer code comprises determining code that is configured to determine that the NNR units, that refer to the NNR model parameter set unit, are independently decodable based on the syntax element.
17. The system of claim 16 , wherein the syntax element is included in a header of the NNR model parameter set unit.
18. The system of claim 11 , wherein the NNR layer parameter set unit includes a syntax element that indicates that NNR units, that refer to the NNR layer parameter set unit, are independently decodable, and the computer code comprises determining code that is configured to determine that the NNR units, that refer to the NNR layer parameter set unit, are independently decodable based on the syntax element.
19. The system of claim 18 , wherein the syntax element is included in a header of the NNR layer parameter set unit.
20. A non-transitory computer-readable medium storing computer instructions that, when executed by at least one processor, cause the at least one processor to:
encode an independent neural network with a topology such as to obtain a neural network representation (NNR) bitstream; and
send the NNR bitstream to a decoder,
wherein the NNR bitstream includes a group of NNR units (GON) that represents the independent neural network with the topology, and the GON includes an NNR model parameter set unit, an NNR layer parameter set unit, an NNR topology unit, an NNR quantization unit, and an NNR compressed data unit, and
each of the NNR model parameter set unit, the NNR layer parameter set unit, the NNR topology unit, the NNR quantization unit, and the NNR compressed data unit is a respective at least one NNR unit, of the GON, that each includes a header and a payload.