Picture prediction method, encoder, decoder and computer storage medium
An image prediction method includes: an initial prediction value of a to-be-predicted colour component of a current block is determined; a sample associated with a reference colour component of the current block is determined; side information of the current block is determined according to the sample associated with the reference colour component; the initial prediction value is filtered by using a preset network model and the side information of the current block, to obtain a target prediction value of the to-be-predicted colour component of the current block; and the to-be-predicted colour component of the current block is encoded according to the target prediction value.
1. A picture prediction method, applied to an encoder, the method comprising:
determining an initial prediction value of a to-be-predicted colour component of a current block;
determining a sample associated with a reference colour component of the current block;
determining side information of the current block according to the sample associated with the reference colour component;
filtering the initial prediction value by using a preset network model and the side information of the current block, to obtain a target prediction value of the to-be-predicted colour component of the current block; and
encoding the to-be-predicted colour component of the current block according to the target prediction value.
2. The method of claim 1 , wherein the reference colour component comprises one or more colour components in a current picture that are different from the to-be-predicted colour component, the current picture being a picture where the current block is located.
3. The method of claim 1 , wherein determining the side information of the current block according to the sample associated with the reference colour component comprises:
determining the sample associated with the reference colour component as the side information of the current block.
4. The method of claim 1 , wherein determining the side information of the current block according to the sample associated with the reference colour component comprises:
performing a first filtering processing on the sample associated with the reference colour component according to a colour component sampling format, to obtain a filtered sample associated with the reference colour component; and
determining the filtered sample associated with the reference colour component as the side information of the current block.
5. The method of claim 4 , further comprising:
signalling the colour component sampling format;
determining a value of a bit field to be signalled, wherein the value of the bit field is used for indicating the colour component sampling format; and
signalling the value of the bit field.
6. The method of claim 1 , wherein filtering the initial prediction value by using the preset network model and the side information of the current block, to obtain the target prediction value of the to-be-predicted colour component of the current block comprises:
inputting the side information of the current block and the initial prediction value of the to-be-predicted colour component into the preset network model to output the target prediction value of the to-be-predicted colour component by the preset network model.
7. The method of claim 6 , further comprising:
determining the preset network model,
wherein determining the preset network model comprises:
determining network model parameters; and
constructing the preset network model according to determined network model parameters.
8. The method of claim 7 , wherein the preset network model comprises a neural network model and a first adder,
wherein the neural network model comprises at least one of: a convolution layer, a residual layer, an average pooling layer, or a sampling conversion module, and
wherein the residual layer comprises at least one of: an activation function, a convolution layer, or a second adder.
9. The method of claim 8 , wherein inputting the side information of the current block and the initial prediction value of the to-be-predicted colour component into the preset network model to output the target prediction value of the to-be-predicted colour component by the preset network model comprises:
inputting the side information and the initial prediction value into the neural network model to output an intermediate value; and
performing addition processing on the intermediate value and the initial prediction value by the first adder to obtain the target prediction value.
10. The method of claim 7 , wherein determining the preset network model comprises:
acquiring a training sample set, wherein the training sample set comprises one or more pictures;
constructing an initial network model, and training the initial network model by using the training sample set; and
determining a trained initial network model as the preset network model.
11. A picture prediction method, applied to a decoder, the method comprising:
parsing a bitstream to acquire a target prediction mode of a current block;
determining an initial prediction value of a to-be-predicted colour component of the current block according to the target prediction mode;
determining a sample associated with a reference colour component of the current block;
determining side information of the current block according to the sample associated with the reference colour component;
filtering the initial prediction value by using a preset network model and the side information of the current block, to obtain a target prediction value of the to-be-predicted colour component of the current block; and
decoding the to-be-predicted colour component of the current block according to the target prediction value.
12. The method of claim 11 , wherein the reference colour component comprises one or more colour components in a current picture that are different from the to-be-predicted colour component, the current picture being a picture where the current block is located.
13. The method of claim 11 , wherein determining the side information of the current block according to the sample associated with the reference colour component comprises:
determining the sample associated with the reference colour component as the side information of the current block.
14. The method of claim 11 , wherein determining the side information of the current block according to the sample associated with the reference colour component comprises:
performing a first filtering processing on the sample associated with the reference colour component according to a colour component sampling format, to obtain a filtered sample associated with the reference colour component; and
determining the filtered sample associated with the reference colour component as the side information of the current block.
15. The method of claim 14 , further comprising:
parsing the bitstream to acquire the colour component sampling format;
parsing a data unit of a parameter set in the bitstream to acquire a value of a bit field for indicating the colour component sampling format; and
determining the colour component sampling format according to the value of the bit field.
16. The method of claim 11 , wherein filtering the initial prediction value by using the preset network model and the side information of the current block, to obtain the target prediction value of the to-be-predicted colour component of the current block comprises:
inputting the side information of the current block and the initial prediction value of the to-be-predicted colour component into the preset network model to output the target prediction value of the to-be-predicted colour component by the preset network model.
17. The method of claim 16 , further comprising:
determining the preset network model,
wherein determining the preset network model comprises:
parsing the bitstream to acquire network model parameters of the preset network model; and
determining the preset network model according to determined network model parameters.
18. The method of claim 17 , wherein the preset network model comprises a neural network model and a first adder,
wherein the neural network model comprises at least one of: a convolution layer, a residual layer, an average pooling layer, or a sampling conversion module, and
wherein the residual layer comprises at least one of: an activation function, a convolution layer, or a second adder.
19. The method of claim 18 , wherein inputting the side information of the current block and the initial prediction value of the to-be-predicted colour component into the preset network model to output the target prediction value of the to-be-predicted colour component by the preset network model comprises:
inputting the side information and the initial prediction value into the neural network model to output an intermediate value; and
performing addition processing on the intermediate value and the initial prediction value by the first adder to obtain the target prediction value.
20. The method of claim 17 , wherein determining the preset network model comprises:
acquiring a training sample set, wherein the training sample set comprises one or more pictures;
constructing an initial network model, and training the initial network model by using the training sample set; and
determining a trained initial network model as the preset network model.