Reproduce “LAMP: Layerwise Adaptive Mixing of Predictors for Training-Free Few-Shot Vision-Language Adaptation” 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
Zhipeng Ye, Feng Jiang, Qiufeng Wang, Hao Li
LAMP adapts frozen CLIP for few-shot image classification by fitting separate residual predictors to global and intermediate-layer regional features, without projecting them into a shared space. Validation data selects how their score corrections are combined. At 16 shots across eleven datasets, RN50 achieves 78.73% accuracy versus 76.62% for a global-only control; gains require validation data and add inference cost.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.