You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Own the stamp extraction in-house; propagate the coadd centroid (drop sf_tools) #810
Resolves Axel's condition on #767 (options 2+3): reimplement the postage-stamp extraction in ShapePipe and propagate the coadd centroid at the same time, so the wcs centroid prior is computed in exactly the same way as the stamp extraction — by construction, not by after-the-fact matching.
What changes
vignetmaker gains a shapepipe-owned get_stamps() (zero-pad by radius, slice around round(pos)), replacing sf_tools.image.stamp.FetchStamps. Bit-identical to FetchStamps for in-bounds objects; an out-of-bounds centre now raises instead of being silently wrapped modulo the image shape.
The extractor records, per object per epoch, the rounded extraction pixel and the sub-pixel offset, carried on the multi-epoch vignette (no sidecar). ngmix's wcs (coadd) centroid consumes that stored offset instead of re-projecting the sky position through galsim and re-rounding — so a milli-pixel WCS disagreement can no longer flip a rounding tie and put the centroid prior a full pixel off.
CENTROID_SOURCE hsm|wcs stays selectable. sf_tools is dropped from the dependencies.
Out of scope: the real-data hsm-vs-wcs A/B (needs a real-data run), and removing the HSM path (the open question on #767).
Activity
cailmdaley commentedon Jul 12, 2026
Folding this into #767 directly — PR #811 now closes #767 itself; this intermediate issue just restated the PR. — Claude on behalf of Cail