Hey Warp Fam,
It's been a while.
Most importantly, this version supports the latest Nvidia GPUs because it uses PyTorch for everything ML-related. Unfortunately, this breaks all backward-compatibility with previously trained BoxNet and denoiser models. The latter also includes M species, which you can quickly recreate by reusing the half-maps, mask, and particles from the last iteration of the old species. On the bright side, you can finally use fast GPUs with a lot of memory.
M finally supports Relion 3.1-style STAR files, which should make particle import less annoying. No support for Relion 4.0 – I hope 4.1 finally adopts particle tilt series, which might lead to further format changes soon.
Apart from that, you won't notice many changes. Small tweaks here and there, bug fixes, probably some new bugs.
Still to come in a future beta are command line versions of Warp and M for Linux, and a properly debugged Noise2Tomo, which is an improved take on IsoNet's wonderful idea. I also have yet to write a tutorial for Noise2Tomo, but you might be able to figure it out from the command line options. You need to denoise the tomograms with Noise2Map first, keeping the default folder structure.
I hope to see some of you at the Tomo Congress tomorrow.
Sorry for the delay,
Dimitry