Reproduce “Virtual Task Prompting for Multi-Task Scene Understanding” 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
Zedong Wang, Yucheng Wang, Dan Xu
Virtual Task Prompting lets multi-task vision models share information through a small external state without changing a standard vision transformer. Learned routes read task-specific vectors from this state before each block and write processed vectors back afterward. On Pascal-Context, it reports a +2.77% average per-task relative gain over TaskPrompter, averaging relative changes with lower-is-better signs reversed; gains are not uniform.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.