Reproduce “FlexReader: Towards Adaptive Decoding Paths for Efficient Arbitrary-Length Scene Text Recognition” 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
Zhibin Ma, Pengwen Dai, Zhigang Yao, Xugong Qin, Wenqi Ren, Xiaochun Cao
FlexReader reads words in natural-scene images by predicting short character runs in parallel, then committing only the confident prefix before decoding again. Its shared substring representation supports strides from one to its trained perception length. On Union14M-B’s long-text subset, L=8, APD reaches 78.1% word accuracy at 32.29 ms, versus 72.4% at 193.23 ms for ARM, under the reported test setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.