We commonly receive data where the in-focus slice is at the very edge of the stack (e.g. some over-confluent OPS experiments, cellanome, likely upcoming ipsc work), and waveorder's current 2D phase reconstruction performs poorly.
Here is an example case where (left) the edge-most raw data slice shows the sharpest features while the (center) same slice in the 3D reconstructions is ~equally sharp and the (right) 2D reconstruction suffers from strong ringing. I would expect a one-sided reconstruction to be worse than a two-sided reconstruction, but not this much worse.
Ideas: The current 2D phase reconstruction estimates both phase and absorption...which reduces to looking for axially odd (phase) and even (absoroption) double-cone PSFs. One-sided data can't distinguish axially odd from even PSFs (phase from absorption), so we may need to be smart with our padding or use other tricks to make this problem more invertible.
cc @gav-sturm
We commonly receive data where the in-focus slice is at the very edge of the stack (e.g. some over-confluent OPS experiments, cellanome, likely upcoming ipsc work), and
waveorder's current 2D phase reconstruction performs poorly.Here is an example case where (left) the edge-most raw data slice shows the sharpest features while the (center) same slice in the 3D reconstructions is ~equally sharp and the (right) 2D reconstruction suffers from strong ringing. I would expect a one-sided reconstruction to be worse than a two-sided reconstruction, but not this much worse.
Ideas: The current 2D phase reconstruction estimates both phase and absorption...which reduces to looking for axially odd (phase) and even (absoroption) double-cone PSFs. One-sided data can't distinguish axially odd from even PSFs (phase from absorption), so we may need to be smart with our padding or use other tricks to make this problem more invertible.
cc @gav-sturm