tree d406fcfd3d59b992ddf1a59a27ea54ff8ed20b54
parent 0c7ca36897b1cd8a6d8c4bdf51d125e9dc1dda20
author Rachel Barker <rachelbarker@google.com> 1678747284 +0000
committer Rachel Barker <rachelbarker@google.com> 1678832570 +0000

Fix some minor issues in global motion code

* Fix a check which was supposed to be "is model type IDENTITY or
  TRANSLATION?" but was actually written as "is model type not ROTZOOM?"

* Track ref_frame_error for the best warp model found, and use this when
  deciding whether to use that model. We were accidentally using the
  ref_frame_error for the last model considered instead.

* Also strengthen some checks on when to skip models, in cases where
  they definitely won't be useful.

None of these change the encoder output currently - we currently only
consider one model per ref frame, of type ROTZOOM - but this helps
future-proof things so that we can more accurately evaluate multiple
models or use AFFINE type models in the future.

Change-Id: Ieb7edd200ab99dcc2ba98e42b46c79166c260f79