Loss reduction inconsistency between RetinaNet and FasterRCNN

keras-team / keras-cv

Industry-strength Computer Vision workflows with Keras

Other

1k stars 330 forks source link

Open tanzhenyu opened 1 year ago

tanzhenyu commented 1 year ago

One takes SUM, one takes NONE. We should unify that before 0.4 release

tanzhenyu commented 1 year ago

This has been unified per 0.4 release, however it's not using the AUTO from core Keras. We're currently evaluating whether we should make it or not