diff options
| author | Hui Su <huisu@google.com> | 2018-07-16 22:05:19 -0700 |
|---|---|---|
| committer | Hui Su <huisu@google.com> | 2018-08-31 11:50:01 -0700 |
| commit | 2713dba538406c10dd9684f2df877aef014583d8 (patch) | |
| tree | a282e8765f5566af0b2dbc21ce852a386ab67f30 /vp9/encoder/vp9_context_tree.h | |
| parent | dbcb89be244e03074ee80e7bc6d5c22e6fc5dd79 (diff) | |
| download | libvpx-2713dba538406c10dd9684f2df877aef014583d8.tar libvpx-2713dba538406c10dd9684f2df877aef014583d8.tar.gz libvpx-2713dba538406c10dd9684f2df877aef014583d8.tar.bz2 libvpx-2713dba538406c10dd9684f2df877aef014583d8.zip | |
ML based rectangular partition search pruning
Add a ML model to predict if rectangular partition search can be skipped
without much coding loss. This model is enabled for speed 0 low bitdepth
only.
Impact on coding performance is minor:
avg_psnr ovr_psnr ssim
lowres -0.005% 0.005% 0.017%
midres 0.100% 0.114% 0.134%
hdres 0.048% 0.083% 0.074%
jvet480p 0.035% 0.027% 0.044%
jvet720p 0.094% 0.090% 0.174%
Tested encoding speed over 20 midres and hdres clips, average speed
gain is about 8%; maximum speed gain is 23%.
Change-Id: I5d4029dec7134c53ac68ab6cf0c8077dc0b767ed
Diffstat (limited to 'vp9/encoder/vp9_context_tree.h')
| -rw-r--r-- | vp9/encoder/vp9_context_tree.h | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/vp9/encoder/vp9_context_tree.h b/vp9/encoder/vp9_context_tree.h index 2bcc26e94..c4600d4bc 100644 --- a/vp9/encoder/vp9_context_tree.h +++ b/vp9/encoder/vp9_context_tree.h @@ -56,6 +56,7 @@ typedef struct { // scope of refactoring. int rate; int64_t dist; + int64_t rdcost; #if CONFIG_VP9_TEMPORAL_DENOISING unsigned int newmv_sse; |