Reproduce “Resolution Allocation with Learnable Resizing and Spectral Filter Parameterization” 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
Wei Wu, Eric C. Larson
The authors test how learned resolution allocation interacts with convolutional filter type. They compare standard 3 × 3 spatial filters with SpecMLP spectral filters under matched complexity budgets, using learnable downsampling. Across four cropped-resizing test settings, spectral filters show a 0.013 to 0.028 larger gain in held-out accuracy than spatial filters. This finding is limited to the tested backbone and parameterization; the cause remains unexplained.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.