Reproduce “Déjà-Events: Multi-view 3D Geometry with Event Cameras via Sparsity-Aware Adaptive Gating” 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
Arda Cinar Demirtas, Pablo Varela Bernal, Marc Pollefeys, Sunghwan Hong
Déjà-Events predicts depth, camera pose and 3D pointmaps from event recordings by distilling a frozen RGB geometry model using aligned frames during training. Its patchwise refinement gates adapt to event density while inference uses events alone. On zero-shot DSEC validation, pointmap AbsRel is 0.148 versus 0.301 for EventVGGT A among comparable event-based methods; the RGB teacher scores 0.058.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.