RoMa out there matching things I didn't intend :)

if you’re seriously trying to understand agi, core concepts you should familiarize yourself with: y = ax +b, y = max(x,0), y = x / ||x||, y = f(x) + x

AGI is when LLMs produce correct bibs

IMO a good CUDA implementation of dense corresp RANSAC would fix bottom issue.

This holds up even for any p(doom) > 0

"ah, it's 19 bits then, right" "well not the storage, that's still 32 bit" "oh, so then the accumulation is done in 19 bits then, right" "no, the accumulation is still done in full fp32" "oh, so then there's basically no precision loss from fp32 then" "well, also no"

When people ask me "surely you must be feeling proud and relaxed now that your model is out"

After each deadline

Cool combination of mast3r robustness with roma refiners :)

MuM: Multi-View Masked Image Modeling for 3D Vision David Nordström Johan Edstedt Fredrik Kahl Georg Bökman tl;dr: multiview CroCo. Great for matching (RoMa), good but not best for depth/semseg. arxiv.org/abs/2511.17309