dLux
dLux copied to clipboard
LouisDesdoigts
Differentiable optical models as parameterised neural networks in Jax using Zodiax
The `CoronLayer` should have attrs: `pad`, `layers`, `crop_output` `InterpolatedLayer` should have attrs: `array`? `transformation`, `order`
PR to track changes. Adding the `ApplyAsymmetricJitter` class I wrote for TOLIMAN, parametrised by `r`, `phi`, and `shear`. I basically completely rewrote the `generate_kernel` method in a way that seems...
Update the `generate_kernel` method of the existing class. Perhaps add in the `ApplyLinearJitter` class I wrote for Toliman work. e.g. ```python class ApplyJitter(...): ... def generate_kernel(self: DetectorLayer, pixel_scale: Array) ->...
If wavefront amplitudes are set _upstream_ before the propagation, then the necessarily small values output by unitary PSFs can be avoided. This should help alleviate FP precision issues, but is...
Currently vmap batches the whole computation as one which can result in hitting RAM bottlenecks, resulting in a pretty hard-core slow down. Looks like there are some long-term plans to...
Hi all, The `git` repository currently occupies a quarter of a gigabyte, which seems like it is far too large. After running `du -h .git` I find that the majority...
As part of increased confidence in testing, I think we could add in testing coverage metrics (https://breadcrumbscollector.tech/how-to-use-code-coverage-in-python-with-pytest/) and show them off on the README. This would also help target testing.
I've noticed that a few of the tests are only checking if not nan or inf. I understand that for PSF's etc it can be difficult to hard code test...
Hi all, Inspired by `numpyro` and `ticktack` drawing automatic figures, I was thinking that we could do something similar in `dLux` and draw the optical system using `pytikz`. The following...
Metadata
Owner
Metadata
Differentiable optical models as parameterised neural networks in Jax using Zodiax