Summary

• Fixed _validate_camera_param to expect (ny, nx) instead of (nx, ny) for calibration matrices
• The previous convention was inconsistent with Julia's [row, col] indexing
• This caused BoundsError for rectangular cameras and semantic errors (transposed data) for square cameras

Details

The validation check for calibration matrices (readnoise, gain, offset, qe) was expecting size(param) == (nx, ny), but Julia arrays use A[row, col] indexing. For a camera with nx columns and ny rows, matrices should be (ny, nx).

Bug impact:

• For rectangular cameras: BoundsError when downstream code (like GaussMLE) accesses map[y, x]
• For square cameras: No error, but data semantically transposed - pixel (x,y) gets value meant for (y,x)

Why it wasn't caught:

• Most tests used square cameras where (nx, ny) == (ny, nx)
• SMLMData is a container that doesn't access matrix elements itself

Test plan

• Existing tests pass (fixed the one rectangular test that used wrong convention)
• Added explicit test for rectangular camera verifying semantic access map[y, x]
• Added test that transposed matrix throws DimensionMismatch

🤖 Generated with Claude Code