tree a3354538129332abca8a0fec16a768f91cbb2323
parent e61bff9b97344e1020ad2801fd2fed71091a0f3a
author Alexander Bokov <alexanderbokov@google.com> 1536179430 -0700
committer Hui Su <huisu@google.com> 1536599171 +0000

Update TX type pruning models

Add models for 4x16 and 16x4 transforms, retrain models
for all other transform sizes except for 16x16. Remove
no longer necessary models for 32x16 and 16x32 (as only
2 TX types are allowed for these sizes).

Average speed-up on 13 lowres and midres clips (speed=0):
QP = 50 - 1.7%
QP = 40 - 2.0%
QP = 30 - 4.6%

Coding efficiency impact on speed 0:
         avg_psnr  ovr_psnr    ssim
lowres    0.001%    -0.067%   0.006%
midres    0.023%     0.023%   0.022%

STATS_CHANGED

Change-Id: I7f7729b78e0d2bbdf999e91c05e8e37cc96730e9