Hastings @hgreer@qoto.org

I blogged about using the Z3 theorem prover to draw the new aperiodic “Hat” tile: https://www.hgreer.com/HatTile/

@hannorein

@gpowerf I'd lean towards minecraft for that job, but that's probably because I'm a boomer

@waded Ideally, you shouldn't have to fix a website in preferences. If, nonetheless, you want to fix the website, you can make it a stop jiggling with right bar -> scroll down -> preferences -> Site Theme to "Mastodon Light" -> check "slow mode" -> check "reduce animations"

@rgs For me it renders differently every time I open the page, which I have decided just adds to the indie web charm.

@tiago_j_m@sigmoid.social Evaluating who is winning at a game of Risk

@War_Kittens LaTeX for advanced smileys \(^o\nabla^o\)

@StarkeyStardust@scicomm.xyz Here's a shoutout for the team behind Voxelmorph and Synthmorph for writing and publishing excellent code that makes it easy to reproduce their results! @adalca

@StarkeyStardust@scicomm.xyz I'm working on making sure the rival methods that we compare our new method to in an upcoming paper are set up to get the best performance possible. It's always a slightly demoralizing step, since the better job I do, the worse our new method looks in comparison, but at least it's educational to dive into what other people have tried!

The anxiety of having a really good idea to improve your baseline model

@bridgetobarrett Go Heels!

@danbirman We'll get back to it after the CVPR submission deadline and sfn are over

@danbirman Hi! I work in image registration, and I’ve always struggled to visualize the deformations that my code produces. Grid lines work for 2-D slices, but I’m on the lookout for a technique to render the 3-D map nicely.

I was having some trouble with glitchy scrolling using the qoto web interface on mac, especially on the federated timeline. Switching to the default mastodon theme (preferences > appearance > site theme) fixed it immediately!

Agenda for after the CVPR deadline:

Try first step translation transform

Try By-Construction-Inverse-Consistency instead of a penalty- use https://openaccess.thecvf.com/content_cvpr_2016_workshops/w15/papers/Ferraris_Accurate_Small_Deformation_CVPR_2016_paper.pdf for precise SVF integration

Try SVF with inverse consistency penalty, for taste

Try TV regularization instead of GradICON

Try training on diverse corpus of augmented medical images*LNCC + SynthMorph shapes * DICE - use paired t1-t2 medical images when we have them

#vagueposting #science

@davintxi This whole channel is awesome- would recommend

I guess I'm here now instead of twitter. I'm a phd student researching deep learning for medical image registration. If for some awful reason you need to align several hundred CT scans, check out my library icon_registration (https://icon.readthedocs.io)

@djinnome If you already have an automatic differentiation package in mind, it should be able to differentiate through https://github.com/timothyrenner/RungeKutta.jl (very simple, just a 70 line implementation of rk4). I'm not sure about options for more fully featured solvers