Inter prediction method and apparatus based on history-based motion vector
An image decoding method according to the present disclosure includes deriving a history-based motion vector prediction (HMVP) buffer for a current block, deriving motion information of the current block based on an HMVP candidate included in the HMVP buffer, generating prediction samples for the current block based on the motion information, and generating reconstructed samples based on the prediction samples, and the HMVP buffer is initialized when processing a first-ordered CTU in a CTU row in which a current CTU is located, and the current block is included in the current CTU.
1. An image decoding method performed by a decoding apparatus, the method comprising:
deriving a history-based motion vector prediction (HMVP) buffer for a current block;
deriving merge candidates for configuring a merge candidate list for the current block, wherein the merge candidates includes spatial candidates derived based on spatial candidate blocks and a temporal candidate derived based on a temporal candidate block, wherein an HMVP candidate comprised in the HMVP buffer is inserted as a merge candidate of the merge candidate list, wherein the HMVP candidate is inserted after the temporal candidate in the merge candidate list;
deriving motion information of the current block based on the HMVP candidate in the merge candidate list;
generating prediction samples for the current block based on the motion information; and
generating reconstructed samples based on the prediction samples,
wherein the HMVP candidate among the merge candidates comprised in the merge candidate list is indicated based on a merge index obtained from a bitstream,
wherein the current block corresponds to a coding unit (CU) which is split from a coding tree unit (CTU), and
wherein the HMVP buffer is initialized at a first-ordered CTU in a CTU row of a slice.
2. The image decoding method of claim 1 , wherein if the HMVP buffer is initialized, the number of HMVP candidates comprised in the HMVP buffer is set to zero.
3. The image decoding method of claim 1 , further comprising determining whether a current CTU is the first-ordered CTU in the CTU row of the slice.
4. The image decoding method of claim 3 , wherein if it is determined that the current CTU is the first-ordered CTU in the CTU row of the slice, the HMVP buffer comprises an HMVP candidate derived based on motion information of a block decoded earlier than the current block in the current CTU, and
wherein if it is determined that the current CTU is not the first-ordered CTU in the CTU row of the slice, the HMVP buffer comprises an HMVP candidate derived based on motion information of a block decoded earlier than the current block in the current CTU and an HMVP candidate derived based on motion information of a block previously decoded in a previous CTU in the CTU row.
5. The image decoding method of claim 3 , wherein it is determined, based on the following equation, whether the current CTU is the first-ordered CTU in the CTU row of the slice,
ctu_idx % Num
where the ctu_idx refers to an index of the current CTU, the % refers to modulus operator, and the Num refers to the number of CTUs in the CTU row.
6. An image encoding method performed by an encoding apparatus, the method comprising:
deriving a history-based motion vector prediction (HMVP) buffer for a current block;
deriving merge candidates for configuring a merge candidate list for the current block, wherein the merge candidates includes spatial candidates derived based on spatial candidate blocks and a temporal candidate derived based on a temporal candidate block, wherein an HMVP candidate comprised in the HMVP buffer is inserted as a merge candidate of the merge candidate list, wherein the HMVP candidate is inserted after the temporal candidate in the merge candidate list;
deriving motion information of the current block based on the HMVP candidate in the merge candidate list;
generating prediction samples for the current block based on the motion information;
deriving residual samples based on the prediction samples; and
encoding image information comprising information on the residual samples,
wherein the motion information include a merge index indicating the HMVP candidate among the merge candidates comprised in the merge candidate list,
wherein the current block corresponds to a coding unit (CU) which is split from a coding tree unit (CTU), and
wherein the HMVP buffer is initialized at a first-ordered CTU in a CTU row of a slice.
7. The image encoding method of claim 6 , wherein if the HMVP buffer is initialized, the number of HMVP candidates comprised in the HMVP buffer is set to zero.
8. The image encoding method of claim 6 , further comprising determining whether a current CTU is the first-ordered CTU in the CTU row of the slice.
9. The image encoding method of claim 8 , wherein if it is determined that the current CTU is the first-ordered CTU in the CTU row of the slice, the HMVP buffer comprises an HMVP candidate derived based on motion information of a block encoded earlier than the current block in the current CTU, and
wherein if it is determined that the current CTU is not the first-ordered CTU in the CTU row of the slice, the HMVP buffer comprises an HMVP candidate derived based on motion information of a block encoded earlier than the current block in the current CTU and an HMVP candidate derived based on motion information of a block previously encoded in a previous CTU in the CTU row.
10. The image encoding method of claim 8 , wherein it is determined, based on the following equation, whether the current CTU is the first-ordered CTU in the CTU row of the slice,
ctu_idx % Num
where the ctu_idx refers to an index of the current CTU, the % refers to modulus operator, and the Num refers to the number of CTUs in the CTU row.
11. A non-transitory computer-readable storage medium storing encoded information causing a decoding apparatus to perform an image decoding method, the method comprising:
deriving a history-based motion vector prediction (HMVP) buffer for a current block;
deriving merge candidates for configuring a merge candidate list for the current block, wherein the merge candidates includes spatial candidates derived based on spatial candidate blocks and a temporal candidate derived based on a temporal candidate block, wherein an HMVP candidate comprised in the HMVP buffer is inserted as a merge candidate of the merge candidate list, wherein the HMVP candidate is inserted after the temporal candidate in the merge candidate list;
deriving motion information of the current block based on an HMVP candidate in the merge candidate list;
generating prediction samples for the current block based on the motion information; and
generating reconstructed samples based on the prediction samples,
wherein the HMVP candidate among the merge candidates comprised in the merge candidate list is indicated based on a merge index obtained from a bitstream,
wherein the current block corresponds to a coding unit (CU) which is split from a coding tree unit (CTU), and
wherein the HMVP buffer is initialized at a first-ordered CTU in a CTU row of a slice.