Method and apparatus for unification based coding for neural network model compression
A method of unification based coding for neural network model compression is performed by at least one processor and includes receiving a layer uniform flag indicating whether a quantized weight of an input neural network is encoded using a uniform coding method, and determining whether the quantized weight is encoded using the uniform coding method, based on the received layer uniform flag. The method further includes, based on the quantized weight being determined to be encoded using the uniform coding method, encoding the quantized weight, using the uniform coding method, and based on the quantized weight being determined to not be encoded using the uniform coding method, encoding the quantized weight, using a non-uniform coding method.
1. A method of unification based coding for neural network model compression, the method being performed by at least one processor, and the method comprising:
receiving a layer uniform flag indicating whether a quantized weight of an input neural network is encoded using a uniform coding method;
determining whether the quantized weight is encoded using the uniform coding method, based on the received layer uniform flag;
based on the quantized weight being determined to be encoded using the uniform coding method, encoding the quantized weight using the uniform coding method; and
based on the quantized weight being determined to not be encoded using the uniform coding method, encoding the quantized weight using a non-uniform coding method.
2. The method of claim 1 , wherein the encoding the quantized weight, using the uniform coding method, comprises:
partitioning the quantized weight of the input neural network into a plurality of three-dimensional coding tree units (CTU3Ds);
receiving a CTU3D uniform flag indicating whether a quantized weight of one of the plurality of CTU3Ds is encoded using the uniform coding method;
determining whether the quantized weight of the one of the plurality of CTU3Ds is encoded using the uniform coding method, based on the received CTU3D uniform flag;
based on the quantized weight of the one of the plurality of CTU3Ds being determined to be encoded using the uniform coding method, encoding the quantized weight of the one of the plurality of CTU3Ds, using the uniform coding method; and
based on the quantized weight of the one of the plurality of CTU3Ds being determined to not be encoded using the uniform coding method, encoding the quantized weight of the one of the plurality of CTU3Ds, using the non-uniform coding method.
3. The method of claim 2 , wherein the encoding the quantized weight of the one of the plurality of CTU3Ds, using the uniform coding method, comprises encoding the quantized weight of the one of the plurality of CTU3Ds, using a 3D-Unitree coding method.
4. The method of claim 3 , further comprising, based on the quantized weight of the one of the plurality of CTU3Ds being encoded, using the 3D-Unitree coding method, encoding a sign flag indicating whether the quantized weight of the one of the plurality of CTU3Ds is positive or negative.
5. The method of claim 3 , further comprising, based on a quantized weight of each of the plurality of CTU3Ds being encoded, using the 3D-Unitree coding method, encoding a sign flag indicating whether the quantized weight of each of the plurality of CTU3Ds is positive or negative.
6. The method of claim 2 , wherein the encoding the quantized weight of the one of the plurality of CTU3Ds, using the non-uniform coding method, comprises encoding the quantized weight of the one of the plurality of CTU3Ds, using a 3D-Tagtree coding method.
7. The method of claim 6 , further comprising, based on the quantized weight of the one of the plurality of CTU3Ds being encoded, using the 3D-Tagtree coding method, encoding a sign flag indicating whether the quantized weight of the one of the plurality of CTU3Ds is positive or negative.
8. The method of claim 6 , further comprising, based on a quantized weight of each of the plurality of CTU3Ds being encoded, using the 3D-Tagtree coding method, encoding a sign flag indicating whether the quantized weight of each of the plurality of CTU3Ds is positive or negative.
9. An apparatus for unification based coding for neural network model compression, the apparatus comprising:
at least one memory configured to store program code; and
at least one processor configured to read the program code and operate as instructed by the program code, the program code comprising:
receiving code configured to cause the at least one processor to receive a layer uniform flag indicating whether a quantized weight of an input neural network is encoded using a uniform coding method;
determining code configured to cause the at least one processor to determine whether the quantized weight is encoded using the uniform coding method, based on the received layer uniform flag;
first encoding code configured to cause the at least one processor to, based on the quantized weight being determined to be encoded using the uniform coding method, encode the quantized weight, using the uniform coding method; and
second encoding code configured to cause the at least one processor to, based on the quantized weight being determined to not be encoded using the uniform coding method, encode the quantized weight, using a non-uniform coding method.
10. The apparatus of claim 9 , wherein the program code further comprises partitioning code configured to cause the at least one processor to partition the quantized weight of the input neural network into a plurality of three-dimensional coding tree units (CTU3Ds),
wherein the receiving code is further configured to cause the at least one processor to receive a CTU3D uniform flag indicating whether a quantized weight of one of the plurality of CTU3Ds is encoded using the uniform coding method,
the determining code is further configured to cause the at least one processor to determine whether the quantized weight of the one of the plurality of CTU3Ds is encoded using the uniform coding method, based on the received CTU3D uniform flag,
the first encoding code is further configured to cause the at least one processor to, based on the quantized weight of the one of the plurality of CTU3Ds being determined to be encoded using the uniform coding method, encode the quantized weight of the one of the plurality of CTU3Ds, using the uniform coding method, and
the second encoding code is further configured to cause the at least one processor to, based on the quantized weight of the one of the plurality of CTU3Ds being determined to not be encoded using the uniform coding method, encode the quantized weight of the one of the plurality of CTU3Ds, using the non-uniform coding method.
11. The apparatus of claim 10 , wherein the first encoding code is further configured to cause the at least one processor to encode the quantized weight of the one of the plurality of CTU3Ds, using a 3D-Unitree coding method.
12. The apparatus of claim 11 , wherein the first encoding code is further configured to cause the at least one processor to, based on the quantized weight of the one of the plurality of CTU3Ds being encoded, using the 3D-Unitree coding method, encode a sign flag indicating whether the quantized weight of the one of the plurality of CTU3Ds is positive or negative.
13. The apparatus of claim 11 , wherein the first encoding code is further configured to cause the at least one processor to, based on a quantized weight of each of the plurality of CTU3Ds being encoded, using the 3D-Unitree coding method, encode a sign flag indicating whether the quantized weight of each of the plurality of CTU3Ds is positive or negative.
14. The apparatus of claim 10 , the second encoding code is further configured to cause the at least one processor to encode the quantized weight of the one of the plurality of CTU3Ds, using a 3D-Tagtree coding method.
15. The apparatus of claim 14 , wherein the second encoding code is further configured to cause the at least one processor to, based on the quantized weight of the one of the plurality of CTU3Ds being encoded, using the 3D-Tagtree coding method, encode a sign flag indicating whether the quantized weight of the one of the plurality of CTU3Ds is positive or negative.
16. The apparatus of claim 14 , wherein the second encoding code is further configured to cause the at least one processor to, based on a quantized weight of each of the plurality of CTU3Ds being encoded, using the 3D-Tagtree coding method, encode a sign flag indicating whether the quantized weight of each of the plurality of CTU3Ds is positive or negative.
17. A non-transitory computer-readable storage medium storing instructions that, when executed by at least one processor for unification based coding for neural network model compression, cause the at least one processor to:
receive a layer uniform flag indicating whether a quantized weight of an input neural network is encoded using a uniform coding method;
determine whether the quantized weight is encoded using the uniform coding method, based on the received layer uniform flag;
based on the quantized weight being determined to be encoded using the uniform coding method, encode the quantized weight, using the uniform coding method; and
based on the quantized weight being determined to not be encoded using the uniform coding method, encode the quantized weight, using a non-uniform coding method.
18. The non-transitory computer-readable storage medium of claim 17 , wherein the instructions, when executed by the at least one processor, further cause the at least one processor to:
partition the quantized weight of the input neural network into a plurality of three-dimensional coding tree units (CTU3Ds);
receive a CTU3D uniform flag indicating whether a quantized weight of one of the plurality of CTU3Ds is encoded using the uniform coding method;
determine whether the quantized weight of the one of the plurality of CTU3Ds is encoded using the uniform coding method, based on the received CTU3D uniform flag;
based on the quantized weight of the one of the plurality of CTU3Ds being determined to be encoded using the uniform coding method, encode the quantized weight of the one of the plurality of CTU3Ds, using the uniform coding method; and
based on the quantized weight of the one of the plurality of CTU3Ds being determined to not be encoded using the uniform coding method, encode the quantized weight of the one of the plurality of CTU3Ds, using the non-uniform coding method.
19. The non-transitory computer-readable storage medium of claim 18 , wherein the instructions, when executed by the at least one processor, further cause the at least one processor to encode the quantized weight of the one of the plurality of CTU3Ds, using a 3D-Unitree coding method.
20. The non-transitory computer-readable storage medium of claim 18 , wherein the instructions, when executed by the at least one processor, further cause the at least one processor to encode the quantized weight of the one of the plurality of CTU3Ds, using a 3D-Tagtree coding method.