Fix a bug in dice loss when there exist ignored pixels

[Johnson-Wang]

Motivation

Fix a bug in dice loss when there exist ignored pixels. Say we have pred=[0, 0, 0], binary mask=[0, 0, 1], valid_mask=[0, 0, 0], meaning that all labels are ignored. The expected dice loss should be 0. However, the original code incorrectly outputs dice loss = 0.5.

Checklist

• Pre-commit or other linting tools are used to fix the potential lint issues.
• The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.