tree 34378479dfb774fcc2761d52cca6b9a374ee31c6
parent e9ad475c7238a69d4d0e2e360ce0669253a8be85
author Jingning Han <jingning@google.com> 1593625328 -0700
committer Jingning Han <jingning@google.com> 1593707539 +0000

Fix quantization step in tpl model

This improves the compression efficiency for 8-bit input and 10-bit
internal. The coding results in speed 1 show -0.24% gains in lowres
and -0.21% gains in midres.

STATS_CHANGED

Change-Id: I9fdf9395ea05bc0b111dbcc9df164e1d88e2c9c6