Reproduce “A note on data-efficient finetuning of vision foundation models” 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
Niclas Hergenröther, Antonio Orvieto
The paper evaluates visual prompting (VP) for adapting pretrained classifiers across architectures and compares it with linear probing, full fine-tuning, and LoRA. On DTD, VP achieves high training accuracy but lower test accuracy; LoRA outperforms full fine-tuning on five of six base-size models. AutoVP’s fully connected label mapping drives gains in a VMamba-T Flowers102 ablation, not establishing a general rule.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.