Reproduce “REBASE: Reference-Background Subspace Elimination for Training-Free In-Context Segmentation” 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
Mantha Sai Gopal, Jaison Saji Chacko, Harsh Nandwana, Sandesh Hegde, Debarshi Banerjee, Uma Mahesh
REBASE enables one-shot in-context segmentation by removing feature patterns associated with a reference image’s background before matching its marked target to a query image. It projects features away from dominant background directions, then uses similarity prompts and a dense prior to guide frozen SAM. On PACO-Part, adding REBASE raises mean intersection-over-union from 34.87% to 39.28%; it ranks second on PASCAL-Part.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.