Skip to content

Conversation

@arunkannawadi
Copy link
Member

No description provided.

arunkannawadi and others added 6 commits March 19, 2025 15:08
Having NaN values in the weight that multiplies images is bad, since a
single bad image can ruin a pixel regardless of other inputs. Since we
are setting its corresponding weight to be zero, we kill its image (and
variance) values also by multiplying by zero.
In case the variance plane has zeros resulting in inf values in weight,
that needs to be cleared out as well. Instead of mapping only NaN values
to zero, we mapping non-finite values to zero more generally.
It is inconsistent to set quantities to zero, and not guard against
division by the same quantities. Right before the division, we map zeros
to inf, so the image and variance values of those pixels are zeros. We
also set NO_DATA mask bit when that happens.
This unit test mimics a condition that triggered DM-48920.
These additional changes address more of the divide by zero and NaN warnings
@arunkannawadi
Copy link
Member Author

Wrong branch.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants