4:2:0 packing of feature maps

1 . A method of decoding frames from a bitstream, the method comprising:

decoding, from the bitstream, a frame including a first two-dimensional array and a second two-dimensional array having a size smaller than the first two-dimensional array;

producing, from the first two-dimensional array included in the frame, a first tensor having a plurality of feature maps of a first size; and

producing, from the second two-dimensional array included in the frame, a second tensor having a plurality of feature maps of a second size, wherein the second size is smaller than the first size.

2 . The method according to claim 1 , wherein, the first two-dimensional array corresponds to a luma channel and the second two-dimensional array corresponds to a chroma channel.

3 . The method according to claim 1 , wherein, the first tensor and the second tensor are to be processed by a neural network.

4 . The method according to claim 1 , wherein, the first two-dimensional array includes first areas each of which corresponds to a feature map,

wherein the second two-dimensional array includes second areas each of which corresponds to a feature map.

5 . The method according to claim 4 , wherein, positions of at least part of the first areas relatively correspond to positions of at least part of the second areas in the second two-dimensional array.

6 . The method according to claim 1 , wherein, a width of the second two-dimensional array is half of a width of the first two-dimensional array, and

wherein, a heigh of the second two-dimensional array is half of a height of the first two-dimensional array.

7 . An apparatus for decoding frames from a bitstream, the apparatus comprising:

one or more controllers including one or more processors and one or more memories, the one or more controllers configured to:

decode, from the bitstream, a frame including a first two-dimensional array and a second two-dimensional array having a size smaller than the first two-dimensional array;

produce, from the first two-dimensional array included in the frame, a first tensor having a plurality of feature maps of a first size; and

produce, from the second two-dimensional array included in the frame, a second tensor having a plurality of feature maps of a second size, wherein the second size is smaller than the first size.

8 . A non-transitory computer-readable storage medium which stores a program for executing a method of decoding frames from a bitstream, the method comprising:

decode, from the bitstream, a frame including a first two-dimensional array and a second two-dimensional array having a size smaller than the first two-dimensional array;

produce, from the first two-dimensional array included in the frame, a first tensor having a plurality of feature maps of a first size; and

produce, from the second two-dimensional array included in the frame, a second tensor having a plurality of feature maps of a second size, wherein the second size is smaller than the first size.

9 . A method of encoding frames into a bitstream, the method comprising:

obtaining a plurality of tensors including a first tensor having a plurality of feature maps of a first size, and a second tensor having a plurality of feature maps of a second size, wherein the second size is smaller than the first size;

packing the feature maps of the first tensor and the feature maps of the second tensor into a frame having a first two-dimensional array and a second two-dimensional array having a size smaller than the first two-dimensional array, wherein the feature maps of the first size are packed into the first two-dimensional array, and the feature maps of the second size are packed into the second two-dimensional array; and

encoding the frame into the bitstream.

10 . The method according to claim 9 , wherein, the first two-dimensional array corresponds to a luma channel and the second two-dimensional array corresponds to a chroma channel.

11 . The method according to claim 9 , wherein, the first tensor and the second tensor are tensors that are produced by processing a neural network.

12 . The method according to claim 9 , wherein, the first two-dimensional array includes first areas each of which corresponds to a feature map,

wherein, the second two-dimensional array includes second areas each of which corresponds to a feature map.

13 . The method according to claim 12 , wherein, positions of at least part of the first areas relatively correspond to positions of at least part of the second areas in the second two-dimensional array.

14 . The method according to claim 9 , wherein, a width of the second two-dimensional array is half of a width of the first two-dimensional array, and

wherein, a height of the second two-dimensional array is half of a height of the first two-dimensional array.

15 . An apparatus for encoding feature maps into encoded data, the apparatus comprising:

one or more controllers including one or more processors and one or more memories, the one or more controllers configured to:

obtain a plurality of tensors including a first tensor having a plurality of feature maps of a first size, and a second tensor having a plurality of feature maps of a second size, wherein the second size is smaller than the first size;

pack the feature maps of the first tensor and the feature maps of the second tensor into a frame having a first two-dimensional array and a second two-dimensional array having a size smaller than the first two-dimensional array, wherein the feature maps of the first size are packed into the first two-dimensional array, and the feature maps of the second size are packed into the second two-dimensional array; and

encode the frame into the bitstream.

16 . A non-transitory computer-readable storage medium which stores a program for executing a method of encoding frames into a bitstream, the method comprising:

obtaining a plurality of tensors including a first tensor having a plurality of feature maps of a first size, and a second tensor having a plurality of feature maps of a second size, wherein the second size is smaller than the first size;

packing the feature maps of the first tensor and the feature maps of the second tensor into a frame having a first two-dimensional array and a second two-dimensional array having a size smaller than the first two-dimensional array, wherein the feature maps of the first size are packed into the first two-dimensional array, and the feature maps of the second size are packed into the second two-dimensional array; and

encoding the frame into the bitstream.