Reproduce “Loss re-scaling VQA: Revisiting the LanguagePrior Problem from a Class-imbalance View” 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
Yangyang Guo, Liqiang Nie, Zhiyong Cheng, Qi Tian, Min Zhang
The authors propose reweighting answer losses in visual question answering by answer frequency within question types, giving sparse answers more weight and frequent answers less. On VQA-CP v2, UpDn with Soft-CE scores 40.49% without reweighting and 55.93% with reweighting alone. This result supports gains on that benchmark, not proof that language bias is eliminated or visual reasoning established.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.