Reproduce “RuCLIP -- new models and experiments: a technical report” 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
Alex Shonenkov, Andrey Kuznetsov, Denis Dimitrov, Tatyana Shavrina, Daniil Chesakov, Anastasia Maltseva, Alena Fenogenova, Igor Pavlov and 5 more
The authors introduce six ruCLIP variants for matching Russian class descriptions directly with images, avoiding English translation in zero-shot classification. They encode images and class descriptions in a shared space, then select the closest match. In 18 datasets, ruCLIP beats CLIP with OPUS-MT on 11; on Stanford Cars, one variant scores 0.659 versus 0.567. The authors report lower quality than original CLIP.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.