Reproduce “Steerable Transformers for Volumetric Data” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Soumyabrata Kundu, Risi Kondor
The authors introduce a steerable transformer for classification and segmentation, combining steerable-convolution features with attention over distant locations and positional encodings designed to preserve rotational and translational equivariance. Across four benchmark datasets, transformer configurations report higher scores than steerable-convolution comparisons. On BraTS, whole-tumor Dice is 76.37 ± 0.46 versus 74.43 ± 0.77; this comparison does not isolate attention’s effect.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.