Methods and apparatus for reducing the coding latency of decoder-side motion refinement
Embodiments of video coding systems and methods are described for reducing coding latency introduced by decoder-side motion vector refinement (DMVR). In one example, two non-refined motion vectors are identified for coding of a first block of samples (e.g. a first coding unit) using bi-prediction. One or both of the non-refined motion vectors are used to predict motion information for a second block of samples (e.g. a second coding unit). The two non-refined motion vectors are refined using DMVR, and the refined motion vectors are used to generate a prediction signal of the first block of samples. Such embodiments allow the second block of samples to be coded substantially in parallel with the first block without waiting for completion of DMVR on the first block. In additional embodiments, optical-flow-based techniques are described for motion vector refinement.
1. A method comprising:
at a first block, refining a first non-refined motion vector and a second non-refined motion vector to generate a first refined motion vector and a second refined motion vector;
using one or both of the first non-refined motion vector and the second non-refined motion vector, predicting motion information of a second block, the second block being a spatial neighbor of the first block;
predicting the first block with bi-prediction using the first refined motion vector and the second refined motion vector; and
determining a deblocking filter strength for the first block based at least in part on the first non-refined motion vector and the second non-refined motion vector.
2. The method of claim 1 wherein refining of the first non-refined motion vector and the second non-refined motion vector is performed using decoder-side motion vector refinement (DMVR).
3. The method of claim 1 , wherein refining the first non-refined motion vector and the second non-refined motion vector comprises selecting the first refined motion vector and the second refined motion vector to substantially minimize an error metric.
4. The method of claim 3 , wherein the error metric is a template cost, and wherein refining the first non-refined motion vector and the second non-refined motion vector comprises selecting the first refined motion vector and the second refined motion vector to substantially minimize the template cost with respect to a template signal generated by bi-prediction using the first non-refined motion vector and the second non-refined motion vector.
5. The method of claim 4 , wherein the template cost is a sum of absolute differences.
6. The method of claim 3 , wherein the error metric is an optical flow error metric.
7. The method of claim 1 , further comprising predicting motion information of a third block using at least one of the first refined motion vector and the second refined motion vector, wherein the third block and the first block are collocated blocks in different pictures.
8. The method of claim 7 , wherein predicting motion information of the third block is performed using advanced temporal motion vector prediction (ATMVP).
9. The method of claim 1 , wherein predicting motion information of the second block comprises using spatial advanced motion vector prediction (AMVP).
10. The method of claim 1 , wherein predicting motion information of the second block comprises using at least one of the first non-refined motion vector and the second non-refined motion vector as a spatial merge candidate.
11. The method of claim 1 , wherein predicting the motion information of the second block comprises receiving at least one index identifying the first non-refined motion vector or the second non-refined motion vector.
12. The method of claim 1 , further comprising:
adding a motion vector difference to at least one of the first non-refined motion vector and the second non-refined motion vector to generate at least one reconstructed motion vector; and
generating an inter prediction of the second block with the at least one reconstructed motion vector.
13. The method of claim 1 , further comprising generating an inter prediction of the second block using at least one of the first non-refined motion vector and the second non-refined motion vector.
14. A video coding apparatus comprising a processor configured to perform at least:
at a first block, refining a first non-refined motion vector and a second non-refined motion vector to generate a first refined motion vector and a second refined motion vector;
using one or both of the first non-refined motion vector and the second non-refined motion vector, predicting motion information of a second block, the second block being a spatial neighbor of the first block;
predicting the first block with bi-prediction using the first refined motion vector and the second refined motion vector; and
determining a deblocking filter strength for the first block based at least in part on the first non-refined motion vector and the second non-refined motion vector.
15. The apparatus of claim 14 wherein refining of the first non-refined motion vector and the second non-refined motion vector is performed using decoder-side motion vector refinement (DMVR).
16. The apparatus of claim 14 , wherein refining the first non-refined motion vector and the second non-refined motion vector comprises selecting the first refined motion vector and the second refined motion vector to substantially minimize an error metric.
17. The apparatus of claim 14 , further configured to predict motion information of a third block using at least one of the first refined motion vector and the second refined motion vector, wherein the third block and the first block are collocated blocks in different pictures.
18. The apparatus of claim 14 , wherein predicting motion information of the second block comprises using at least one of the first non-refined motion vector and the second non-refined motion vector as a spatial merge candidate.