Reproduce “Learning Where to Focus: Self-Supervised Multi-Scale ViTs for Histopathology” 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
Anabel Stammer, Valay Bundele, Mehran Hosseinzadeh, Hendrik P.A. Lensch
CRAFT allocates image resolution during self-supervised pretraining: it builds a coarse view, uses attention to select regions, and refines half while retaining context. A symmetric cross-scale loss encourages complementary representations. On CAMELYON16, coarse DINO features reached 81.4% tumor recall versus 69.0% for fine features; in mixed-scale MIL, mean AUC rose from 95.65% to 97.15% at 4.01 GMACs, in that setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.