Skip to content

Latest commit

 

History

History
72 lines (53 loc) · 2.94 KB

TODO.md

File metadata and controls

72 lines (53 loc) · 2.94 KB
Raw

To do

This document is complementary to the issues page on github.

Planned

  • Improve the documentation, and give more examples.

  • Finish dw dots and document it well.

  • Update to the latest VkFFT version.

  • Add a signature to the fimcl_t objects and to validate that the right kind of objects are used in functions using them.

  • Better build instruction for native builds on Windows help wanted

  • Figure out how if it works on Apple silicon help wanted

  • dynamic loading of OpenCL, for examples using [https://github.com/yugr/Implib.so]

  • Complete switch to Vulkan from OpenCL. Doing this will ensure better portability, for example on new macs with MoltenVK. The switch from clFFT to VkFFT in v 0.3.6 was a first step towards this goal.

  • Crash-safe writing of output images. I.e, follow the standard procedure and first, write to temporary file and move to final destination when the write has finished. Avoids some bad luck.

  • Add OpenCL-RL method (as a low mem option).

  • Set default PSF size in dw_bw based on the geometry of the PSF and expected sample thickness.

  • Prepare binary packages, should include Windows 10/11, MacOS and linux.

  • Crop the PSF also in the axial direction when it vanishes, this should be good for confocal, STED etc.

  • --xyfactor is a quite awkward name.

  • Still one kernel to write in order to speedup the initialization in method_shbcl2 when bq > 0.

  • Make the --lookahead option work again, and possibly enable it by default, at least on the CPU side where memory often isn't a limiting factor.

Nice to have

  • Skip libtiff, especially for writing.
  • See if Pinned memory allocations can improve the performance for shbcl (don't expect anything drastic from shbcl2). LFAT, code
  • Wrap malloc to catch allocation errors to the log file?
  • Decide about default tiling settings based on the PSF.
  • Demos, for example on the effect of the tiling.
  • Implement the Gibson-Lanni PSF model.
  • Double check the tiling weights using gradients and other patterns.
  • Add information on the input image to the log -- especially detect saturated pixels.
  • Fail gracefully when a pyramidal tif is supplied.
  • Fail gracefully when a multi color/time/... tif is supplied.
  • Include CCD corrections.
  • Make the background correction work also in tiled mode.
  • Automatic image cropping in the axial direction based on gradient magnitude or similar.
  • Do the padding/unpadding for in-place transformations on the GPU side. This should not affect the memory usage since it is only performed at the beginning and at the end of the algorithms.