Reproduce “Visual Branch is What You Need for CLIP-based Class-Incremental Learning” 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

Tao Hu, Zhen-Hao Xie, Jingcai Guo, De-Chuan Zhan, Da-Wei zhou

VIS is a rehearsal-free method for incremental image classification that builds classifiers from visual features rather than class-name embeddings. It trains a small visual mixer in the base session, then uses a fixed nonlinear feature map and one-versus-all least-squares updates from running statistics, without storing old images. On FGVCAircraft B0 Inc10, VIS reaches 75.21% average and 66.46% final-session accuracy, above BOFA’s 70.96% and 60.43%; findings are limited to tested protocols.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.