Zach Mueller hopes UNet-transformer hybrids put semantic segmentation "back on the menu"
TheZachMueller · x · 2026-11-03
In the thread about classic vision architectures being reinvented, Zach Mueller adds that he'd love to see "unet transformer hybrid nonsense" doing funky things, putting semantic segmentation back on the menu. A concrete wish within the broader observation that classic CV design ideas are returning in new models.
Related event: Classic Vision Architectures Are Being Reinvented from First Principles(3 posts)→
More from Research
- MIT's Soft-MTP: one-line fix removes 80% noisy gradient, boosts MTP pre-training for free — kastnerkyle · 2026-11-03
- Aaron Roth shares calibrated multiaccuracy paper with Jiuyao Lu and Georgy Noarov — Aaroth · 2026-11-03
- Sample Complexity of Calibrated Multiaccuracy Depends on Test-Time Randomization — Aaroth · 2026-11-03
- Calibrated multiaccuracy shows sharp randomized-vs-deterministic sample complexity gap — Aaroth · 2026-11-03
- Multicalibration is often overkill: calibrated multiaccuracy needs fewer samples — Aaroth · 2026-11-03
- MOPD becomes the post-training recipe of the year: TRL now ships multi-teacher on-policy distillation — SergioPaniego · 2026-11-03