Method, an apparatus and a computer program product for video encoding and video decoding

1. A method, comprising:

compressing input data by means of at least a neural network;

determining a compression rate for data compression;

running the neural network with the input data to produce an output data;

removing a number of elements from the output data according to the compression rate to generate a reduced form of the output data;

wherein the compression rate is associated with a mask, based on which the number of elements are removed; and

providing the reduced form of the output data and the compression rate to a decoder.

2. The method according to claim 1 , wherein the neural network is trained with multiple masks associated with different compression rates.

3. The method according to claim 2 , wherein each mask of the multiple masks include a pattern of N ones, and wherein N is different for each compression rate.

4. The method according to claim 3 , wherein first N elements of a dimension of the reduced form of the output data are equal to one and the rest of that dimension are equal to zero.

5. The method according to claim 2 , further comprising:

selecting N randomly from a uniform distribution during training.

6. The method according to claim 2 , further comprising:

selecting N randomly from a non-uniform distribution emphasizing different compression rates or ranges of compression rates.

7. The method according to claim 2 , further comprising:

training the neural network with different values of N, and averaging losses associated with the different values of N for a weight update.

8. A method comprising:

receiving input data for decompression;

decompressing the input data by means of at least a neural network;

determining a decompression rate for decompressing the input data;

running the neural network with the input data to produce a decompressed output data;

padding a number of elements to the decompressed output data according to the decompression rate to produce an output data;

wherein the decompression rate is associated with a mask, based on which the number of elements are padded; and

providing the output data.

9. An apparatus comprising:

at least one processor; and

at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to:

compress input data, wherein compression comprises at least a neural network;

determine a compression rate for data compression;

run the neural network with the input data to produce an output data;

remove a number of elements from the output data according to the compression rate to generate a reduced form of the output data;

wherein the compression rate is associated with a mask, based on which the number of elements are removed; and

provide the reduced form of the output data and the compression rate to a decoder.

10. The apparatus according to claim 9 , wherein the neural network is trained with multiple masks associated with different compression rates.

11. The apparatus according to claim 10 , wherein each mask of the multiple masks include a pattern of N ones, and wherein N is different for each compression rate.

12. The apparatus according to claim 11 , wherein first N elements of a dimension of the reduced form of the output data are equal to one and the rest of that dimension are equal to zero.

13. The apparatus according to claim 10 , wherein the at least one memory and the instructions are further configured to, with the at least one processor, further cause the apparatus to: select N randomly from a uniform distribution during training.

14. The apparatus according to claim 10 , wherein the at least one memory and the instructions are further configured to, with the at least one processor, further cause the apparatus to: select N randomly from a non-uniform distribution emphasizing different compression rates or ranges of compression rates.

15. The apparatus according to claim 10 , wherein the at least one memory and the instructions are further configured to, with the at least one processor, further cause the apparatus to: train the neural network with different values of N, and average losses associated with the different values of N for a weight update.

16. The apparatus according to claim 10 , wherein the at least one memory and the instructions are further configured to, with the at least one processor, further cause the apparatus to: train the neural network with different values of N, and weight the losses based on an inverse frequency of elements a code that are not removed during training.

17. The apparatus according to claim 10 , wherein the at least one memory and the instructions are further configured to, with the at least one processor, further cause the apparatus to: train the neural network with different values of N and have at least one fixed weight based on a value of N.

18. An apparatus comprising:

at least one processor; and

at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to:

receive input data for decompression;

decompress input data, wherein decompression comprises of at least a neural network;

determine a decompression rate for data decompression;

run the neural network with the input data to produce a decompressed output data;

pad a number of elements to the decompressed output data according to the decompression rate to produce an output data;

wherein the decompression rate is associated with a mask, based on which the number of elements are padded; and

provide the output data.