Reproduce “Attention Attention Everywhere: Monocular Depth Prediction with Skip Attention” 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
Ashutosh Agarwal, Chetan Arora
PixelFormer improves monocular depth estimation by refining coarse, scene-aware pixel queries against higher-resolution encoder features through window-based cross-attention. On NYUv2, it reports an absolute relative error of 0.090, compared with 0.095 for NeWCRFs, under a 10-meter cap. In a Swin-T KITTI ablation, the skip-attention module also outperforms tested convolutional fusion alternatives; this evidence is limited to those settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.