Motion candidate derivation with search order and coding mode
Example implementations include methods, apparatuses and computer storage media for video processing. One example method comprises determining, during a conversion between a current video block of a video and a bitstream of the video, a motion candidate for the current video block; refining the motion candidate by performing a local search around the motion candidate based on a template matching cost rule; and performing the conversion based on the refined motion candidate, wherein a search order of the local search is related to a coding mode of the current video block.
1. A video processing method, comprising:
determining, during a conversion between a current video block of a video and a bitstream of the video, a motion candidate for the current video block;
refining the motion candidate by performing a local search around the motion candidate based on a template matching cost rule; and
performing the conversion based on the refined motion candidate,
wherein a search order of the local search is related to a coding mode of the current video block, and wherein the search order specifies that the local search is performed from coarse granularity to fine granularity, and wherein if the coding mode is a merge mode, the search order of the local search is allowed to be integer sample offset searching followed by half sample offset searching followed by quarter sample offset searching followed by one-eighth sample offset searching followed by one-sixteenth sample offset searching,
wherein the integer sample offset searching comprises a first search round with a diamond search pattern and a second search round with a cross search pattern, and wherein each of the half sample offset searching, the quarter sample offset searching, the one-eighth sample offset searching, and the one-sixteenth sample offset searching comprises a single search round with the cross search pattern.
2. The video processing method of claim 1 , wherein if the coding mode is a motion vector predication mode, the search order is one of the following:
integer sample offset searching followed by quarter sample offset searching followed by one-sixteenth sample offset searching;
integer sample offset searching followed by quarter sample offset searching; or
integer sample offset searching followed by half sample offset searching followed by quarter sample offset searching.
3. The video processing method of claim 1 , wherein performing the local search comprises:
determining a starting search point for the local search based on a template of the current video block and motion information indicated by the motion candidate;
performing a first round of integer sample offset searching from the starting search point based on a first search pattern, wherein the first round of the integer sample offset searching is terminated in response to a first search point acting as a center point of the first search pattern and associated with a minimum template matching cost being found;
performing a second round of the integer sample offset searching from the first search point based on a second search pattern, wherein the second round of the integer sample offset searching is terminated in response to a second search point associated with the minimum template matching cost being found; and
determining the refined motion candidate based on the second search point.
4. The video processing method of claim 3 , wherein determine the refined motion candidate based on the second search point comprises:
in response to the local search comprising no fractional sample offset searching, determining the refined motion candidate corresponding to the second search point.
5. The video processing method of claim 3 , wherein determining the refined motion candidate based on the second search point comprises:
in response to the local search comprising fractional sample offset searching, performing the fractional sample offset searching from the second search point based on the second search pattern, to determine a final search point associated with the minimum template matching cost during the fractional sample offset searching; and
determining the refined motion candidate corresponding to the final search point.
6. The video processing method of claim 3 , wherein the first search pattern is an unrestricted center-biased diamond search pattern and the second search pattern is a cross search pattern.
7. The video processing method of claim 1 , wherein a search range of the local search is determined based on a predetermined number of luminance samples.
8. The video processing method of claim 1 , wherein the local search comprises integer sample offset searching and fractional sample offset searching, and performing the local search comprises:
determining an integer sample offset for the integer sample offset searching;
performing the integer sample offset searching based on the integer sample offset;
determining a fractional sample offset for the fractional sample offset searching by using a parametric error surface equation; and
performing the fractional sample offset searching based on a result of the integer sample offset searching and the fractional sample offset.
9. The video processing method of claim 8 , wherein the integer sample offset searching is terminated in response to a search point associated with a minimum template matching cost being found, and determining the fractional sample offset comprises:
using a template matching cost of the search point and respective template matching costs of a plurality of neighboring search points to fit the parametric error surface equation; and
determining the fractional sample offset by solving the parametric error surface equation based on the minimum template matching cost.
10. The video processing method of claim 8 , wherein performing the fractional sample offset searching comprises:
in response to the integer sample offset searching being terminated with a search point associated with a minimum template matching cost being found, performing the fractional sample offset searching to derive the refined motion candidate by adding the fractional sample offset to a motion candidate corresponding to the search point.
11. The video processing method of claim 1 , wherein the video processing method is only performed at CU-level.
12. The video processing method of claim 1 , wherein the video processing method is applied for a Luma component.
13. The video processing method of claim 1 , wherein the video processing method is applied for at least one Chroma component.
14. The video processing method of claim 1 , wherein different template shapes and/or different matching rules are applied for different components.
15. The video processing method of claim 1 , wherein the conversion comprises encoding the current video block into the bitstream.
16. The video processing method of claim 1 , wherein the conversion comprises decoding the current video block from the bitstream.
17. The video processing method of claim 1 , wherein the refined motion candidate being associated with the minimum template matching cost during the local search.
18. The video processing method of claim 1 , wherein each motion candidate comprises predicted motion information determined from neighboring video blocks of the current video block.
19. An apparatus for processing video data comprising a processor and a non-transitory memory with instructions thereon, wherein the instructions upon execution by the processor, cause the processor to:
determine, during a conversion between a current video block of a video and a bitstream of the video, a motion candidate for the current video block;
refine the motion candidate by performing a local search around the motion candidate based on a template matching cost rule; and
perform the conversion based on the refined motion candidate,
wherein a search order of the local search is related to a coding mode of the current video block, and wherein the search order specifies that the local search is performed from coarse granularity to fine granularity, and wherein if the coding mode is a merge mode, the search order of the local search is allowed to be integer sample offset searching followed by half sample offset searching followed by quarter sample offset searching followed by one-eighth sample offset searching followed by one-sixteenth sample offset searching,
wherein the integer sample offset searching comprises a first search round with a diamond search pattern and a second search round with a cross search pattern, and
wherein each of the half sample offset searching, the quarter sample offset searching, the one-eighth sample offset searching, and the one-sixteenth sample offset searching comprises a single search round with the cross search pattern.
20. A non-transitory computer-readable storage medium storing instructions that cause a processor to:
determine, during a conversion between a current video block of a video and a bitstream of the video, a motion candidate for the current video block;
refine the motion candidate by performing a local search around the motion candidate based on a template matching cost rule; and
perform the conversion based on the refined motion candidate,
wherein a search order of the local search is related to a coding mode of the current video block, and wherein the search order specifies that the local search is performed from coarse granularity to fine granularity, and wherein if the coding mode is a merge mode, the search order of the local search is allowed to be integer sample offset searching followed by half sample offset searching followed by quarter sample offset searching followed by one-eighth sample offset searching followed by one-sixteenth sample offset searching,
wherein the integer sample offset searching comprises a first search round with a diamond search pattern and a second search round with a cross search pattern, and
wherein each of the half sample offset searching, the quarter sample offset searching, the one-eighth sample offset searching, and the one-sixteenth sample offset searching comprises a single search round with the cross search pattern.
21. A non-transitory computer-readable recording medium storing a bitstream of a video which is generated by a method performed by a video processing apparatus, wherein the method comprises:
determining, during a conversion between a current video block of a video and a bitstream of the video, a motion candidate for the current video block;
refining the motion candidate by performing a local search around the motion candidate based on a template matching cost rule; and
performing the conversion based on the refined motion candidate,
wherein a search order of the local search is related to a coding mode of the current video block, and wherein the search order specifies that the local search is performed from coarse granularity to fine granularity, and wherein if the coding mode is a merge mode, the search order of the local search is allowed to be integer sample offset searching followed by half sample offset searching followed by quarter sample offset searching followed by one-eighth sample offset searching followed by one-sixteenth sample offset searching,
wherein the integer sample offset searching comprises a first search round with a diamond search pattern and a second search round with a cross search pattern, and
wherein each of the half sample offset searching, the quarter sample offset searching, the one-eighth sample offset searching, and the one-sixteenth sample offset searching comprises a single search round with the cross search pattern.